Fep 2

pubs.acs.
org/JCTC Article
Improved Alchemical Free Energy Calculations with Optimized

Smoothstep Softcore Potentials
Tai-Sung Lee, Zhixiong Lin, Bryce K. Allen, Charles Lin, Brian K. Radak, Yujun Tao, Hsu-Chun Tsai,
Woody Sherman, and Darrin M. York*
Cite This: J. Chem. Theory Comput. 2020, 16, 5512−5525 Read Online
ACCESS Metrics & More Article Recommendations *

sı Supporting Information
Downloaded via UNIVERSIDAD DE LOS ANDES ANDES on November 1, 2022 at 04:02:21 (UTC).
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
ABSTRACT: Progress in the development of GPU-accelerated free energy simulation

software has enabled practical applications on complex biological systems and fueled
efforts to develop more accurate and robust predictive methods. In particular, this work re-
examines concerted (a.k.a., one-step or unified) alchemical transformations commonly
used in the prediction of hydration and relative binding free energies (RBFEs). We first
classify several known challenges in these calculations into three categories: endpoint
catastrophes, particle collapse, and large gradient-jumps. While endpoint catastrophes have
long been addressed using softcore potentials, the remaining two problems occur much
more sporadically and can result in either numerical instability (i.e., complete failure of a simulation) or inconsistent estimation (i.e.,
stochastic convergence to an incorrect result). The particle collapse problem stems from an imbalance in short-range electrostatic
and repulsive interactions and can, in principle, be solved by appropriately balancing the respective softcore parameters. However,
the large gradient-jump problem itself arises from the sensitivity of the free energy to large values of the softcore parameters, as might
be used in trying to solve the particle collapse issue. Often, no satisfactory compromise exists with the existing softcore potential
form. As a framework for solving these problems, we developed a new family of smoothstep softcore (SSC) potentials motivated by
an analysis of the derivatives along the alchemical path. The smoothstep polynomials generalize the monomial functions that are
used in most implementations and provide an additional path-dependent smoothing parameter. The effectiveness of this approach is
demonstrated on simple yet pathological cases that illustrate the three problems outlined. With appropriate parameter selection, we
find that a second-order SSC(2) potential does at least as well as the conventional approach and provides vast improvement in terms
of consistency across all cases. Last, we compare the concerted SSC(2) approach against the gold-standard stepwise (a.k.a.,
decoupled or multistep) scheme over a large set of RBFE calculations as might be encountered in drug discovery.
■ INTRODUCTION
Recent progress and improvements in computer hardware,
critical to obtain stable, converged results with affordable
sampling.
One of the first major obstacles that was encountered in
simulation software, and free energy methods,1−12 especially,
simple concerted (a.k.a., one-step or unified) linear alchemical
the development of highly efficient and cost-effective GPU transformations was the “endpoint catastrophe.”33−39 Numer-
accelerated free energy calculations,12−18 have significantly ical singularity or instability occurs when evaluating ensembles
extended the accessible timescales of computer simulations generated with one Hamiltonian using the potential energy at
and scope of applications. In addition to ongoing challenges of other points along the alchemical pathway (as required to
developing more accurate force fields and efficient sampling obtain thermodynamic derivatives or perturbations) for which
methods, there is need to improve our ability to optimally set there is poor phase space overlap. The most severe case occurs
up alchemical free energy calculations.19−32 with linear interpolation of two real state endpoints, where the
The setup problem refers to not only how to create the difference between the real state potential energy functions is
relevant necessary input files but also the proper simulation required to evaluate the thermodynamic derivative or
protocols and parameters that will yield the best results for a exponentiated energy difference.34−36 Because different atoms
given system of interest. In alchemical free energy simulations, in the two states can be artificially superimposed in such a
one of the most difficult but pivotal technical issues is the
choice of the alchemical path connecting the two real states Received: March 10, 2020
(i.e., connecting the two thermodynamic endpoints). Although Published: July 16, 2020
the free energy difference between two states is independent of
the path that connects them in the regime of complete
conformational sampling, in practical calculations of complex
systems, the choice of the alchemical transformation path is
© 2020 American Chemical Society https://dx.doi.org/10.1021/acs.jctc.0c00237

5512 J. Chem. Theory Comput. 2020, 16, 5512−5525
Journal of Chemical Theory and Computation pubs.acs.org/JCTC Article
transformation, this can lead to energy differences or dynamics,47−50 Hamiltonian replica exchange methods,51−55
thermodynamic derivatives that are unstably large in adaptive biasing,48,56,57 or self-adjusted mixture sampling58,59
magnitude and even singular. The singularity arises due to methods. Consequently, it is of practical interest to work
“hard” (i.e., inverse power law form of the interaction potential toward a more robust and efficient solution for concerted
at short distances) exchange repulsions and/or Coulombic alchemical transformations.
interactions between atoms in the core transformation region In this work, we present methods for improved “concerted”
that unphysically overlap.35,40 alchemical free energy transformations of complex biomolec-
There are two general approaches to address the endpoint ular systems. The remainder of the paper is outlined as follows:
catastrophe. The first is the use of “softcore potentials” with The Theory section develops the mathematical model
separation-shifted scaling,36,38 short-range switching,39 or framework, provides definitions of the three main problems
capping the short-range interactions.41,42 The second is to commonly encountered in concerted transformations, and
use nonlinear mixing of the endpoint potentials.33−35,37,43 presents a systematic set of formulas to facilitate their
Nowadays, combinations of these two approaches are most discussion. We then formulate a new smooth softcore potential
often utilized.38,42,43 A formalism for minimal variance path using smoothstep functions of variable order P that we
based on the standard form of softcore potentials has also been designate “SSC(P).” In the Results section, we demonstrate
reported.44,45 These have been successful strategies to formally how the second-order smoothstep softcore SSC(2) potential
address the classical endpoint catastrophe, as defined in the with optimized parameters is able to overcome all problematic
present context. cases of concerted alchemical transformations we have yet
Nonetheless, with the use of softcore potentials to address encountered. We demonstrate the robustness of the SSC(2)
the endpoint catastrophe, there remains two other major potential compared to a conventional widely used softcore
problems in concerted transformations that can result in potential for a broad range of hydration free energy and
numerical instabilities and poor statistical results.39 The first is relative binding free energy (RBFE) calculations. Results are
the “particle collapse” problem that stems from an imbalance compared to benchmark quality “stepwise” (decharge-vdW-
of softcore Coulomb attraction and exchange repulsions and recharge) free energy calculations. The Discussion section
can lead to artificial minima, where particles from the two places the work into broader context, and the Conclusions
transforming states are on top of one another. The second is section summarizes the main points of the paper and identifies
the “large gradient-jump” problem that arises from sensitivity future research directions. The methods presented here have
of the free energy to large values of the softcore parameters been originally implemented as a modified enhancement of the
sometimes required to balance the softcore Coulomb and GPU-accelerated free energy methods in AMBER18,17 which
exchange interactions. have recently been extensively validated,60 and now are in the
An alternate strategy to circumvent the “particle collapse” recent AMBER20 release.61
and “large gradient-jump” problems altogether is to avoid the
use of a concerted transformation and instead use a “stepwise”
approach, sometimes referred to as “multistep” or “split”
■ THEORY
One can formulate the computation of relative free energies
procedures,46 in which the electrostatic and Lennard-Jones from equilibrium simulations using a thermodynamic pertur-
(LJ) interactions are handled in separate steps in the bation (TP)62 (sometimes referred to as “free energy
alchemical transformation. An example of a stepwise perturbation”) or thermodynamic integration (TI)63,64 or
“decharge-vdW-recharge” strategy would be as follows: through nonequilibrium ensemble simulations using the
• Step 1: Decharge the mutating atoms in the initial state Jarzynski equality and its equation variations.65−70 In the
(with LJ and other parameters fixed at their initial state present work, we focus on equilibrium methods and formulate
values) so that there will be no electrostatic interactions the problem in the TI framework. Nonetheless, the
that can lead to an imbalance in the next step. Note, fundamental barriers to progress that we address herein are
during this step, all interactions including electrostatics not specific to the TI method; as we show through numerical
from the final state are turned off. examples, the new methods presented here are equally
• Step 2: With all mutating atoms in both the initial and transferable to TP methods with Bennett Acceptance Ratio
final states turned off (i.e., “decharged”), transform the (BAR) analysis and its multistate variant (MBAR).71−75
LJ parameters, including r−12 exchange repulsions, from TI with Original AMBER Softcore Potentials. The free
the initial state to the final state, along with any energy is a state function, and thus the free energy difference
additional bonded parameters. between thermodynamic states is independent of the path that
connects them (assuming fully converged sampling along the
• Step 3: Recharge the mutating atoms (with LJ and other path). Computationally, however, the choice of this pathway is
parameters fixed at their final state values) to their final most often of immense importance because for nontrivial
state values. problems, statistical sampling is required not just at the end
Although stepwise approaches have been demonstrated to states but along the pathway itself.
be quite robust, they have several disadvantages. First, the Consider the transformation of a system of N particles in an
procedure is more tedious to set up and can be sensitive to the initial state “0” characterized by potential energy function
choice of atoms in the decharge/recharge region, in some U0(q), where q represents the degrees of freedom of the
cases, leading to intermediate states that have a significantly system (e.g., Cartesian positions of each particle along with any
different net charge. Second, the procedure is more computa- system variables), to a final state “1” characterized by potential
tionally intensive as it requires more steps, each with different energy function U1(q) having the same degrees of freedom.
sampling requirements and statistical error estimates. Third, Let us define a thermodynamic parameter λ that smoothly
the procedure is not well suited for advanced λ-schedule connects these states through a λ-dependent potential U(q; λ)
optimization and enhanced sampling schemes, such as λ such that U(q; 0) = U0(q) and U(q; 1) = U1(q). In this case,
5513 https://dx.doi.org/10.1021/acs.jctc.0c00237
J. Chem. Theory Comput. 2020, 16, 5512−5525
the change in free energy ΔG0→1 = G1 − G0 can be determined There have been many different proposed softcore potential
through the TI formula forms that modify or “soften” the nonbonded interactions. In
M the discussion below of specific interaction potentials, we
1 ∂U (q; λ) ∂U (q; λ) simplify the presentation by only showing the softcore
ΔG0 → 1 = ∫0 dλ
∂λ
≈ ∑ wk
∂λ potential corresponding to one end state. It should be
λ k=1 λk
reminded that the total system potential energy at a particular
(1)
λ value should contain weighted contributions from both end
where the second sum indicates numerical integration over M state potentials as in eq 4.
quadrature points (λk for k = 1, ..., M) with associated weights The LJ and electrostatic interactions for a set of interacting
ÄÅ É
wk. We now discuss specific ways in which U(q; λ) can be
ÅÅi y12 i y6ÑÑÑ
point particles i and j separated by a distance rij are given by
ÅÅjj σij zz j z ÑÑ
constructed. The simplest way to establish a thermodynamic
ULJ(rij) = 4ϵijÅÅÅÅjjj zzz − jjjj zzzz ÑÑÑÑ

connection is to use a linear interpolation between states,
ÅÅj rij z j rij z ÑÑ
σ
ÅÅÇk { k { ÑÑÖ
ij
which we will designate as UL(q; λ)
U L(q; λ) = (1 − λ)U0(q) + λU1(q) = U0(q) + λΔU (q) (6)
(2) and
ji qiqj zyz 1
UC(rij) = jjj z
where ΔU(q) = U1(q) − U0(q). The simple linear alchemical
j 4π ϵ0 zz rij
k {
transformation pathway has the thermodynamic derivative
(7)
∂U L(q; λ)
= U1(q) − U0(q) = ΔU (q)
∂λ (3) where σij and ϵij are the pairwise van der Waals contact
distance and well depth, respectively, and qi and qj are the
Hence, the common energy components that are identical partial charges of particles i and j, respectively.
between U1(q) and U0(q) need not be explicitly considered as The Cartesian derivatives of these energies are straight
the corresponding difference is zero. As has been well forward as ∇iU(rij) = (rij/rij)(dU/drij) = −∇jU(rij), where the
ÄÅ É
established and is discussed in more detail below, however,
6Ñ
ij ϵij yzÅÅÅÅ ij σij yz ij σij yz ÑÑÑÑ
derivatives with respect to rij are given by
the linear alchemical transformation pathway leads to practical
= −4jjjj zzzzÅÅÅÅ12jjjj zzzz − 6jjjj zzzz ÑÑÑÑ

problems that can be partially overcome by the use of so-called 12
j rij zÅÅ j rij z j rij z ÑÑ

dULJ(rij)
“softcore” potentials. These potentials, in the present context
k {ÅÇÅ k { k { ÑÖÑ
and the discussion that follows, apply only to nonbonded (i.e., drij
(8)
LJ and electrostatic) interactions within the nonbonded
cutoff.36,38 All other components of the energy are achieved and
ij qiqj yz 1
= −jjj zz
here through the conventional linear transformation pathway,
j 4π ϵ0 zz r 2
although other nonlinear pathways are also possible through dUC(rij)
k { ij
parameter interpolation.16 Here, we define the original softcore drij (9)
potential transformation pathway38 in AMBER as
These derivatives are programmed in molecular simulation
U SC(q; λ) = (1 − λ)U0SC(q; λ) + λU1SC(q; 1 − λ) software codes in order to derive the normal electrostatic and
= U0SC(q; λ) + λΔU SC(q; λ) (4) van der Waals forces on particles. In order to “soften” these
pairwise interactions with particles contained within the
where ΔUSC(q;λ) ≡ USC1 (q;1 − λ) − U0 (q;λ). Note that, after
SC
selected softcore region, one can modify the effective
adding the softcore modification, the original λ-independent interaction distance by introducing a parametric form for
end state Hamiltonians U0(q) and U1(q) now become λ- separation-shifted scaling with an adjustable parameter. A
dependent USC SC
0 (q;λ) and U1 (q;λ), respectively. The corre- commonly used form of these modifications is36,38
sponding thermodynamic derivative is given by
rijLJ(λ ; α) = [rij n + λασij n]1/ n (10)
∂U SC(q; λ)
ÅÄÅ
= [U1SC(q; 1 − λ) − U0SC(q; λ)]
ÅÅ ij ∂U SC(q; λ) yz
∂λ and
+ ÅÅÅ(1 − λ)jjjj 0 zz
zz
ÅÅ
rijC(λ ; β) = [rij m + λβ]1/ m
ÅÇ k {
(11)
É
ij ∂U1SC(q; 1 − λ) yzÑÑÑÑ
∂λ
where n and m are positive integers and α and β are adjustable
+ λjjj zzÑÑ
zzÑÑ
j
positive semidefinite parameters for the LJ and electrostatic
k {ÑÑÖ
softcore interactions, respectively, with values of zero
∂λ (5) corresponding to no softcore modification for any λ value. In
several molecular simulation software suites, including the
In the current AMBER implementation, a nonbonded cutoff is
default in AMBER, the values of n = 6 and m = 2 are used,
defined, and the softcore potentials are applied as correction
although other values have also been suggested.38 Note that for
terms to the original nonbonded LJ and electrostatic
positive values of α and β parameters and λ values between 0
interactions between the softcore region atoms and other
and 1, the effective interaction distances satisfy the conditions
non-softcore atoms within the nonbonded cutoff. As a result,
when utilizing the particle-mesh Ewald method (PME),76 only rijLJ(λ ; α) ≥ rij (12)
the direct space term evaluated within the nonbonded cutoff is
affected by the softcore potentials. Details are described in the
rijC(λ ; β) ≥ rij (13)
Supporting Information.
where the equality holds only for the real state endpoint λ = 0. often encountered with simple linear alchemical trans-
This results in “softening” of the LJ and electrostatic formations where the total potential is defined as the linear
interactions with increasing values of α and β parameters, combinations of the end state potentials and occurs when the λ
respectively, particularly at short range where they are the value approaches a real state endpoint where the potential
largest in magnitude. From these modified effective interaction energy of one real state endpoint needs to be evaluated with
distances, the LJ and electrostatic softcore potentials can be the conformational ensemble generated from the other real
defined as state endpoint and can result in unphysical atom−atom overlap
SC and thermodynamic derivatives. Equation 3 shows that the
ULJ (rij ; λ) = ULJ[rijLJ(λ ; α)] (14) required thermodynamic derivative involves a difference
between real state endpoint potentials (i.e., ΔU(q)) for all λ
and
points, even if this difference is unstably large. Formally, the
UCSC(rij ; λ) = UC[rijC(λ ; β)] (15) endpoint catastrophe can occur not only at the endpoints of an
alchemical transformation but at any point that produces the
The thermodynamic derivatives with respect to λ can be symptoms as described above.
obtained using the chain relation as Solution. The endpoint catastrophe can be avoided by the
SC use of original AMBER softcore potentials. Formally, the
∂ULJ (rij ; λ) dULJ[rijLJ(λ ; α)] ∂rijLJ(λ ; α) endpoint catastrophe can be avoided by use of the softcore
= ·
∂λ drijLJ(λ ; α) ∂λ (16) potentials presented in eq 4, with particular sensitivity to the α
parameter that tunes the “softness” of the LJ terms, including
where the short-ranged repulsive potential. Larger β and especially α
values thus tend to have the greatest affect in alleviating the
∂rijLJ(λ ; α) endpoint catastrophe.
= (α /n) ·σij n·[rijLJ(λ ; α)](1 − n)/ n
∂λ (17) Particle Collapse. Definition. With softcore potentials,
particle collapse is the artificial superposition of particles at
and
intermediate values of λ that can lead to large amplitude
∂UCSC(rij ; λ) dUC[rijC(λ ; β)] ∂rijC(λ ; β) fluctuations or phase transition behavior along the λ
= · dimension.42
∂λ drijC(λ ; β) ∂λ (18) Cause. Particle collapse results from an imbalance of
Coulomb attraction and exchange repulsions that favor atomic
where
overlap. Although softcore potentials formally eliminate
∂rijC(λ ; β) singularities at the endpoints, they can also lead to the
= (β /m) ·[rijC(λ ; β)](1 − m)/ m creation of new artificial minima most commonly at very small
∂λ (19)
interaction distances (r → 0, but possibly at other locations for
Problems with Original AMBER Softcore Potentials complex interactions) if the short-range exchange repulsion
and Their Underlying Causes. The field of free energy terms of the softcore LJ potential are not sufficient to
methods is vast, and yet, there currently exists no commonly overcome the softcore Coulomb attractions of oppositely
accepted and consistently used terminology that enables charged particles.38,39 This could cause unwanted alchemical
discussion of certain classes of problems that can occur when traps or even phase transition-like behavior along the λ
performing free energy simulations. Here, we define the main dimension, resulting in unstable, unconverged numerical
problems that we have encountered regarding the alchemical results.
transformation path in free energy simulations and have Solution. The Coulomb-exchange imbalance problem can
endeavored to overcome in the current work. Further, we be overcome by adjustment (decrease) of the α/β softcore
distinguish the problems themselves, which manifest as parameter ratio and/or modification of the softcore potential
observed symptoms in the simulations, from their underlying functional form. In order to avoid Coulomb-exchange
origins (causes), which we relate mathematically to the imbalance, the softcore exchange repulsions need to be made
equations in order to motivate their solutions. “harder” (smaller α value) and/or the softcore Coulomb
There are three main problems related to the alchemical interactions need to be made “softer” (larger β value). While
path and the associated softcore potential that can commonly this adjustment is somewhat system-dependent, as will be
occur in free energy simulations and, in particular, with so- illustrated below, certain combinations appear to be remark-
called “concerted transformations” that involve simultaneous ably robust for a variety of transformations in different
changes in both nonbonded LJ and electrostatic terms. We will environments for a range of protein−ligand systems.
refer to these as the endpoint catastrophe, the particle collapse Large Gradient-Jump. Definition. With softcore potentials,
problem, and the large gradient-jump problem. We now the sensitivity of the free energy for large values of the softcore
discuss each of these in more detail, including defining the parameters can lead to spurious jumps in the free energy near
symptoms of each problem, highlighting their underlying the thermodynamic endpoints.
causes, and outlining the solutions. Cause. Large jumps in free energy can result from sensitivity
Endpoint Catastrophe. Definition. With linear alchemical of the thermodynamic derivatives (gradient) to certain softcore
transformations, endpoint catastrophe is the sharp divergence parameter values near the real state endpoints. This is
(and sometimes singularity) of the free energy difference or the particularly manifested when large β values are required to
thermodynamic derivative at the thermodynamic endpoints (λ adjust the α/β softcore parameter ratio to solve the Coulomb-
values near 0 and 1). exchange imbalance problem. This can result in a dramatic
Cause. Endpoint catastrophes are due to poor phase space increase in the terms in eq 5 involving λ derivatives of USC1 and
overlap. The endpoint catastrophe is a well known problem USC0 at the λ = 1 and 0 endpoints, respectively. Equations 17
and 19 show that the thermodynamic derivatives for the ∂U SC[q; SP(λ)]
softcore LJ and Coulomb interactions are roughly linear in the = {U1SC[q; 1 − SP(λ)] − U0SC[q; SP(λ)]}
l
o
∂SP(λ)
+o
parameters α and β, respectively, particularly for λ values near
m
o
o
the real state endpoints where energy changes are typically the ∂U0SC[q; SP(λ)]
n
largest and nonlinear.39 [ 1 − SP ( λ ) ]·
∂U [q; 1 − SP(λ)] |
∂SP(λ)
o
o
Solution. The large gradient-jump problem can be solved by
}
o
o
formulating a more sophisticated smooth softcore potential SC
~
with derivatives that vanish at the endpoints. We formulate a + SP(λ) · 1
family of smooth softcore potentials that use smoothstep ∂SP(λ) (25)
weighting functions with favorable endpoint derivative proper-
ties with the intent of developing and testing a robust softcore Note that eq 23 for USSC(P)(q; λ) is identical to eq 4 for USC(q;
framework that can be used for efficient concerted trans- λ) with the latter having λ argument replaced by SP(λ).
formations. Further, note that USSC(0)(q; λ) (P = 0 family member) is
With proper choice of α and β softcore parameters, together identical to USC(q; λ). Thus, the original softcore potential is
with an appropriately smooth weighting function (see below), contained as the lowest order member of the SSCpotential
we find a solution across all above problems for a wide range of family described here. However, this is the only member of the
alchemical transformations, including several severe problem- family that does not have derivative values that vanish at the
atic test cases. boundaries (λ = 0 and 1) but rather has a constant value of 1
Formulation of a Family of Smoothstep Softcore over the range of λ. As we will demonstrate below, other
Potentials for Robust Concerted Alchemical Trans- higher-order members of the family have numerical properties
formations. We consider a family of smoothstep functions, that are more well-behaved, and, in particular, the S2 family
SP(λ), of orders P (P = 0, 1, 2, ...) and defined as the member, with appropriate choice of α and β values, overcomes
polynomial functions (up to P = 4 shown) all problematic cases of concerted transformations. As
mentioned earlier, the SSC potential introduced here is only
for 0 ≤ x ≤ 1: applied to nonbonded LJ and electrostatic interactions
S0(x) = x , between atoms in the softcore region and atoms in the non-
softcore region within the cutoff. To be consistent, the
S1(x) = −2x 3 + 3x 2 , smoothstep combination scheme of potentials (the prefactors
1 − SP(λ) and SP(λ) in eq 23) is applied to not only the above
S2(x) = 6x 5 − 15x 4 + 10x 3 , softcore terms but also the corresponding long-range analytic
dispersion “tail” corrections77 beyond the cutoff and 1−4
S3(x) = −20x 7 + 70x 6 − 84x 5 + 35x 4 , scaled nonbonded terms. Other components of the λ-
S4(x) = 70x 9 − 315x 8 + 540x 7 − 420x 6 + 126x 5 , dependent potential energy without softcore potential
correction and their thermodynamic derivatives do not use
and the smoothstep functions but use the original linear
combination scheme. These terms include the PME reciprocal
SP(x < 0) = 0; SP(x > 1) = 1, ∀P∈
space term (including the “self-energy” and net charge
(20) correction), bonded terms, and restraint terms.
■
The smoothstep functions are monotonically increasing
functions that have the desirable endpoint values METHODS
SP(0) = 0; SP(1) = 1 ∀ P ∈  (21) Simulation Setup and Protocols. A modified version of
AMBER18 with the proposed SSC(P) scheme implementa-
ÅÄÅ k ÑÉ ÅÄÅ k ÑÉ
ÅÅ d SP(x) ÑÑÑ ÅÅ d SP(x) ÑÑÑ
and derivative properties tion, now in AMBER20,61 was employed for all simulations. All
ÅÅ ÑÑ = ÅÅ Å ÑÑ
ÅÅ Ñ Ñ
simulations were performed with the recently implemented
ÅÅÇ dx k ÑÑÑÖx = 0 ÅÅÅÇ dx k ÑÑÑÖx = 1

= 0 ∀ k ∈ , 0 < k ≤ P GPU-TI modules14,17 built against the CUDA 10.1 GPU
library and run on various GPU workstations and servers
(22) equipped with NVIDIA GTX 1080TI, RTX 2080 TI, Titan V,
and V100 GPUs. Results reported were created with the single
From these smoothstep functions, we create a family of precision calculation/flexible precision accumulation model.78
smooth softcore potentials for nonbonded LJ and electrostatic The setups and protocols for AMBER standard GPU-
interactions involving atoms in the softcore region by replacing accelerated TI simulations14,17 are employed. The AMBER
λ with SP(λ) in eqs 4 and 5 to obtain ff14SB force field79 is used for standard amino acids and
U SSC(P)(q; λ) = U SC[q; SP(λ)] GAFF80 with the AM1-BCC charges81,82 for molecules without
AMBER parameters. The TIP3P water model83 is utilized. The
= [1 − SP(λ)]·U0SC[q; SP(λ)] parmed module of AMBER18 is used to prepare the topology
files for TI calculations. The SHAKE algorithm84,85 is used to
+ SP(λ) ·U1SC[q; 1 − SP(λ)] (23) constrain bonds between heavy atoms and hydrogens, except
with thermodynamic derivatives in the mutating parts, where no SHAKE is applied. Long-range
electrostatic interactions are treated by the PME method.76 A
∂U SSC(P)(q; λ) ∂U SC[q; SP(λ)] dSP(λ) cutoff of 10 Å for the model systems (see below sections) and
= · the hydration free energy calculations and a cutoff of 8 Å for
∂λ ∂SP(λ) dλ (24)
RBFEs are used for nonbonded interactions, including the
where, following from eq 5, we have direct space terms of the PME method and particles interacting
Figure 1. ⟨∂U/∂λ⟩λ vs λ plots for alchemical simulations of three molecular systems using the one-step concerted scheme: the absolute hydration
free energies for diphenyl toluene (upper rows, denoted as DPT/0) and single Na+ ion (lower rows, denoted as Na+/0) and the relative hydration
free energy simulations for the Factor Xa ligand L51c to L51h mutation (middle rows, denoted as L51c/h). The L51c ligand has 65 atoms, and
L51h has 58 atoms. The red-colored atoms shown are the defined softcore regions, that is, the unique atoms for the individual ligands. The atoms
common to both ligands are not shown, except the connecting carbon shown in black. The left three columns show the result using the original
AMBER softcore potentials. These three columns show different β values (Å2), and different colored curves correspond to different α values
(unitless). The rightmost column shows the results from the proposed SSC(2) with the optimal softcore parameters (α = 0.2, β = 50 Å2). Each
curve represents one 101-window (total 5 ns) TI simulation, and there are four simulations for each condition. Note that the endpoint and the large
gradient-jump problems with the original softcore potentials near λ = 0 and 1 are absent in the results with the SSC(2) potential. The particle
collapse problem shown in L51c/h around λ = 0.2 to 0.4 and 0.7 to 0.8 and in Na+/0 around λ = 0.2 with the original softcore potential disappears
with the SSC(2) potential.
through softcore potentials. The Ewald error tolerance is set to we demonstrate that the SSC(2) potential can reproduce
10−5, and the Ewald coefficient is automatically set according benchmark stepwise transformations for a broad range of
to the error tolerance and nonbonded cutoff. The model RBFEs of ligand sets for eight protein targets.
systems are simulated with the NVT ensemble at 300 K, Illustration of Problems Using Original AMBER
whereas the hydration free energy and the RBFE simulations Softcore Potentials and Proposed Solutions. In this
are performed with the NPT ensemble regulated at 1.0 atm at section, we examine a set of simple edge cases that illustrate
298 K. Detailed setups and protocols for individual systems the endpoint catastrophe, particle collapse, and large gradient-
can be found in the Supporting Information. jump problems and go on to propose a solution using a new
■ RESULTS
Here, we develop a P-order smoothstep potential, SSC(P), to
second-order SSC potential, SSC(2), with optimized param-
eters. The test cases involving absolute and relative hydration
free energies (gas phase part of the cycle is not considered in
overcome problems inherent in concerted transformations with these illustrations) are as follows: The first test case involves
the conventional softcore potential38 currently used in many calculation of the absolute hydration free energy of 3,4-
molecular simulation packages. Although many values of P diphenyltoluene (denoted as the DPT/0), a bulky fairly
have been explored in this work, P = 2 (second order) was hydrophobic system, which will be made to vanish in solution.
found to have the most reliable performance and hence will be The second test case is the absolute hydration free energy of a
the focus in the following results and discussion. Nevertheless, Na+ ion (denoted Na+/0), a small charged system that will
higher orders of P may be useful for specific scenarios not introduce new issues when made to vanish in solution. The
encountered in this work, and therefore, this variable has been third test system is the relative hydration free energy of two
retained for the user to adjust, with P = 2 being the default in Factor Xa ligands,14,86 L51c and L51h, involving the
AMBER20. We begin by examining a select set of alchemical transformation L51c → L51h in solution (denoted as L51c/
transformations in solution that represent edge cases to h) and migration of charge from one region of the ligand to
illustrate the origin of the endpoint catastrophe, particle another.
collapse, and large gradient-jump problems and how these Results of the Original AMBER Softcore Potentials.
problems can be overcome by use of the SSC(2) potential with Figure 1 shows the ⟨∂U/∂λ⟩λ versus λ plots for alchemical free
optimized parameters. We first demonstrate below how energy simulations of these test systems using the one-step
SSC(2) can robustly handle concerted alchemical trans- concerted scheme and different α and β softcore parameters.
formations for a variety of absolute and relative hydration The DPT/0 (upper panels), L51c/h (middle panels), and
free energies of molecules in comparison with benchmark Na+/0 (lower panels) transformations are shown in solution.
stepwise (decharge-vdW-recharge) transformations. Next, we The original AMBER softcore potential form of eq 4 is used, or
consider a more complex set of alchemical transformations equivalently, the zeroth order SSC function, SSC(0), of eq 23,
involving protein−ligand binding that represent edge cases in the results shown in the first three left columns, whereas the
where the conventional softcore potential fails (sometimes results from the proposed SSC(2) are shown in the rightmost
dramatically) for concerted transformations and demonstrate column. In eq 25, one could also apply the smooth step
that the SSC(2) potential overcomes these problems. Finally, function only to the weights of the Hamiltonians, that is, only
Table 1. Comparison of Relative Hydration Free Energiesa Obtained by Stepwise and Concerted Protocols with the Original
AMBER Softcore Potentials (Labeled as “Original”) and with the Proposed Second-Order Smoothstep Function (Labeled as
SSC(2))
stepwise concerted
transformation original original Δb SSC(2) Δb
methane → 0 −2.34(02) −2.34(02) 0.00 −2.36(03) 0.02
methanol → 0 3.78(02) 3.84(03) 0.06 3.85(04) 0.07
ethane → 0 −2.51(02) −2.54(03) 0.03 −2.54(04) 0.03
toluene → 0 0.84(04) 0.80(04) 0.04 0.82(07) 0.02
neopentane → 0 −2.66(04) −2.67(07) 0.01 −2.69(06) 0.03
2-methylfuran → 0 0.57(03) 0.56(04) 0.01 0.56(06) 0.01
2-methylindole → 0 6.25(04) 6.26(04) 0.01 6.22(08) 0.03
2-cyclopentanylindole → 0 6.59(05) 6.56(05) 0.03 6.55(09) 0.04
7-cyclopentanylindole → 0 6.85(06) 6.78(05) 0.07 6.73(10) 0.12
Avg. Δ 0.03 0.04
methane → ethane 0.07(06) 0.07(08) 0.00 0.04(08) 0.03
methanol → methane 6.19(06) 6.26(07) 0.07 6.21(07) 0.02
methanol → ethane 6.19(04) 6.87(06) 0.68 6.21(04) 0.02
toluene → methane 3.23(07) 3.24(09) 0.01 3.25(10) 0.02
methane → neopentanec 0.07(14) 0.00(18) 0.07 −0.01(18) 0.08
methane → neopentaned 0.23(07) 0.20(10) 0.03 0.21(10) 0.02
2-methylfuran → methane 2.90(07) 2.95(09) 0.05 2.95(10) 0.05
2-methylindole → methane 8.66(07) 8.74(10) 0.08 8.66(12) 0.00
7-CPIe→ 2-CPIf 0.04(11) 0.08(14) 0.04 0.19(16) 0.15
Avg. Δ 0.11 0.04
a
All free energy results were ΔΔG obtained by TI, except that the data from the stepwise scheme with traditional softcore potential are obtained by
BAR. Free energy simulations with the original AMBER softcore potential used the parameters α = 0.5 and β = 12 Å2 that are the default in
AMBER18, and simulations with the SSC(2) potential used the parameters α = 0.2 and β = 50 Å2 developed and tested here that are the default in
AMBER20. The upper rows show the absolute hydration ΔΔG values, whereas the lower rows show the relative hydration ΔΔG values. Both are
obtained by ΔΔG = ΔGaq − ΔGgas. bUsing the results from stepwise scheme, traditional softcore potential with BAR as the reference to show the
errors with respect to the reference. cCentral mapping. dTerminal mapping. e7-cyclopentanylindole. f2-cyclopentanylindole.
replace λ with SP(λ) in the prefactors of eq 25, not the λ of Na+/0, the strong electrostatic interactions require a β value
function argument within the softcore potentials. This of 50 Å2 to achieve stable (smooth) ⟨∂U/∂λ⟩λ curves for α
“midway” scheme applied to the weights only reduces the values of 0.05 and 0.2.
endpoint catastrophe but does not eliminate the large gradient- However, the large value of β required to address the particle
jump problem and is compared with the original softcore collapse problem also has the effect of leading to much larger
scheme and SSC(2) with different α/β parameters in Figure S4 values of ⟨∂U/∂λ⟩λ that arise from the derivative (gradient)
of the Supporting Information. term in eq 19 that is scaled by β. In the Na+/0 transformation,
Consider first the leftmost panels in Figure 1, corresponding this manifests as a dramatic rise in ⟨∂U/∂λ⟩λ at λ = 0 for larger
to a β value of 12 Å2, which is the default value in AMBER18. β values. This is the large gradient-jump problem, which has
For this value of β, with small α values (0.05 and 0.2), the also been discussed by others.39
endpoint “catastrophe” is formally averted, but its effects are At this point, we conclude that the default softcore potential
still clearly prevalent for DPT/0 (large negative ⟨∂U/∂λ⟩λ at λ parameters α/β in AMBER18, 0.5/12 Å2, are useful to
= 1) and L51c/h (large positive ⟨∂U/∂λ⟩λ at λ = 0 and large adequately address the “endpoint catastrophe” for the edge
negative at λ = 1), whereas it is much less apparent for Na+/0 cases considered here but have considerable susceptibility to
which involves much smaller steric annihilation. the “particle collapse” problem. The particle collapse problem
Softening the repulsive potential by using a larger value of α can be addressed by reducing α to harden the short-ranged
(0.5) reduces the problems at the endpoints but leads to other repulsions of the softcore LJ potential and increasing β to
issues at intermediate states. Specifically, for L51c/h, the soften the electrostatic interactions. Nonetheless, with this
profiles are not smooth for λ values between 0.2 and 0.4 and λ strategy for choice of α and β, using the conventional linear
∼ 0.8, and for Na+/0, for λ ∼ 0.2. The origin of this irregularity softcore scheme of eq 4, it does not appear possible to
is the particle collapse problem, where at some intermediate λ simultaneously address the endpoint catastrophe, particle
values, the softened exchange repulsions (large α) can no collapse, and large gradient-jump problems, arising from
longer counterbalance the attractive softcore Coulomb small values of α and large values of β.
attractions of oppositely charged particles, causing them to Development of a Second-Order SSC Potential. We
collapse on top of one another and leading to sampling issues. now consider an alternative SSC function, which employs a
This imbalance can be reduced by increasing the value of β, second-order smoothstep function, SSC(2), as described in eq
as indicated by the second leftmost (β = 17 Å2) and third 23. This form of softcore potential was tested and
leftmost (β = 50 Å2) columns of panels in Figure 1, which demonstrated to conserve energy as well as the regular GPU-
softens the electrostatic interactions. With β set to 17 Å2, the accelerated MD of the real state endpoint and not exacerbate
⟨∂U/∂λ⟩λ curve for L51c/h is much improved relative to β = energy drift. The rightmost column of Figure 1 shows the ⟨∂U/
12 Å2, particularly for the α values of 0.05 and 0.2. In the case ∂λ⟩λ versuss λ plots for alchemical free energy simulations of
the same systems as Figure 1 but with the SSC(2) softcore several relative hydration free energies. Results from
potential as opposed to the original AMBER softcore potential. simulations of concerted transformations with the original
To be clear, only results with reoptimized softcore parameters AMBER softcore potential and SSC(2), with the set are
(α = 0.2, β = 50 Å2) are shown here. The results using the compared with reference calculations using the stepwise
SSC(2) potential with other combinations of α/β can be found approach with original AMBER softcore potential and BAR71
in Figure S5 of the Supporting Information. analysis. Free energy simulations with the original AMBER
The first thing that is apparent is that the behavior at all of softcore potential used the parameters α = 0.5 and β = 12 Å2
the λ = 0 and 1 endpoints has greatly improved. This is that are the default in AMBER18, and simulations with the
because the derivatives of the smoothstep function vanish at SSC(2) potential used the parameters α = 0.2 and β = 50 Å2
the endpoints, making the first term in brackets in eq 24 also developed and tested here that are the default in AMBER20. In
go to zero at λ = 0 and 1. In fact, it is clear from eq 24 that the most cases, the differences between the free energy values with
conventional thermodynamic derivative of the softcore respect to the reference results, designated as Δ in the table,
potential is multiplied by a term that involves the derivative are comparable to or less than the error estimates of the results
of the smoothstep function, which is zero at the endpoints for themselves. A minor exception occurs for the absolute
all orders greater than zero (in which case the derivative is dehydration free energy of methanol (3.78 kcal/mol), which
unity). This also has the consequence that the “large gradient- is slightly overestimated with both the original AMBER
jump” problem due to large values of β at the endpoints is also softcore potential and SSC(2) by 0.06−0.07 kcal/mol. The
mitigated. largest deviation with respect to the reference values occurs for
The use of the smoothstep function is not guaranteed to the methanol → ethane transformation, which is overestimated
resolve the particle collapse problem for all possible cases. The by 0.68 kcal/mol with the original AMBER softcore potential,
particle collapse problem, in its most general form, arises from whereas SSC(2) agrees closely (0.02 kcal/mol). The only
the creation of new artificial minima along the λ alchemical instance where SSC(2) performs statistically more poorly than
progress coordinate. It is therefore possible that use of the the original AMBER softcore potential is for the 7-CPI → 2-
SSC, for a particular set of α/β softcore parameters, in some CPI transformation, which is overestimated by 0.15 kcal/mol
cases may only shift the artificial minima to different regions of with SSC(2), whereas the original AMBER softcore potential is
the q, λ space. Nonetheless, what the smoothstep function closer to the reference result (0.04 kcal/mol). Overall, the
does allow is the stable adjustment of the α/β softcore error estimates for SSC(2) appear very slightly larger than the
parameters so as to mitigate particle collapse without the original AMBER softcore potential, and overall, the results are
adverse consequences of the endpoint and large gradient-jump
quite comparable, except in the instances of the edge cases
problems that occur when the conventional softcore potential
where SSC(2) provides accurate results, whereas the original
is used (see Figures S4 and S5 in the Supporting Information).
AMBER softcore potential fails.
Note that, from eq 25, the total softcore contributions to ⟨∂U/
RBFE Calculation of eight Published Protein Systems.
∂λ⟩λ will be exactly zero at the end points, but the ⟨∂U/∂λ⟩λ
In this section, we examine more complex cases of RBFE on a
results shown in Figure 1 are not zero because of other
series of previously studied drug targets.88 This set (defined as
interaction terms, which currently are not treated with the
proposed SSC scheme. the Wang et al. dataset) covers 8 protein systems and 314
For α/β values of 0.2/50 Å2, all of the edge cases examined ligand mutations and has been widely used as a benchmark test
here appear quite stable. Higher-order smoothstep functions set for tractable RBFE calculations without significant
approach the endpoints more gradually and smoothly as a conformational changes or other challenging scenarios such
consequence of having a steeper slope in the transition as changes in tautomer/ionization states or buried waters. We
between 0 and 1 in the intermediate λ values (Figure S4 of the first examine two systems that demonstrate significant edge
Supporting Information). Comparison of smoothstep func- cases where the original AMBER softcore potential fails,
tions of different orders indicated a good balance between whereas the SSC(2) potential is shown to be accurate. Next,
these properties to use a second-order smoothstep function we examine the six additional protein−ligand systems to
(see Supporting Information for details). In the remainder of demonstrate that the SSC(2) potential is also robust.
the paper, we will use the SSC(2) potential together with α/β Results of the Two Problematic Protein Targets:
values of 0.2/50 Å2 to test and validate against a broad range of PTP1B and p38. Out of the Wang et al. dataset, two protein
hydration free energy and RBFE calculations. Comparison will targets (PTP1B and p38) demonstrated significant edge cases
be made with the original AMBER softcore potential with where the conventional softcore potential with default
default α/β values of 0.5/12 Å2. These methods have recently parameters was observed to fail. The RBFE predictions for
been extensively validated60 in AMBER18 and the new PTP1B ligands are plotted in Figure 2, comparing the stepwise
AMBER20 release.61 scheme (x-axis) and the concerted scheme (y-axis). The upper
Hydration Free Energies of Small Organic Molecules. two panels show the results for the original softcore potential
Here, we examine absolute and relative hydration free energies parameter set analyzed by TI (left panels) and MBAR (right
of a series of small organic molecules that have been recently panels). The lower two panels show the corresponding results
employed to verify the reproducibility of free energy for the SSC(2) potential. It is clear that the original AMBER
calculations across different molecular simulation software softcore potential produces several prominent outlier points
packages.87 The purpose is to verify that the SSC(2) potential, with respect to the stepwise scheme. These outliers are
with parameters adjusted by consideration of the edge cases in exacerbated for the TI results, which are more sensitive to the
the previous section, is robust in reproducing results from the integration of thermodynamic derivatives. Nonetheless, the
stepwise scheme. Table 1 lists the free energy values for MBAR results are clearly problematic with the original softcore
alchemical transformations representing the absolute (de)- potential. With the use of the SSC(2) potential, the RBFE
hydration free energies for nine organic molecules as well as results agree with those of the stepwise scheme to within
close agreement (within statistical error estimates, typically less

than 1 kcal/mol) with the reference values. Further, the TI and
MBAR results are almost indistinguishable. Hence, it appears
that the SSC(2) potential can successfully overcome
limitations of the original AMBER softcore potential for the
problematic edge cases considered here, where the main causes
are the particle collapse problem according to the ⟨∂U/∂λ⟩λ
curves [see typical problematic ⟨∂U/∂λ⟩λ curves in the
Supporting Information].
Results of the Six Well-Behaved Protein Targets. The
RBFE results for the remaining six targets of the Wang et al.
dataset using the concerted scheme are shown in Figure 4 for
Figure 2. RBFE results for PTP1B ligands: Upper panels: simulation

results with the original AMBER softcore potential parameter set and
the linear combination of the softcore potentials, analyzed by TI (left
panels) and MBAR (right panels). Lower panels: simulation results
with the proposed second-order smoothstep function SSC(2) and the
softcore potential parameter set (α = 0.2, β = 50 Å2), analyzed by TI
(left panels) and MBAR (right panels). Plots show the comparison
between the stepwise scheme (x-axis) and the concerted scheme (y-
axis). The dashed red lines indicate the region of ±1.0 kcal/mol
difference. The calculated |ϵ|s are positive definite values and not
normally distributed; hence their standard deviations (shown in
parentheses) should not be interpreted as indicative of data range.
statistical error estimates, and the TI and MBAR results are

virtually identical. Figure 4. Results (y-axis) of RBFE for six targets of the Wang et al.
Analogous results for p38 ligands are plotted in Figure 3. dataset using the concerted scheme and the AMBER original softcore
Again, with the original AMBER softcore potential, both TI potential with default parameters (α = 0.5, β = 12 Å2) are compared
and MBAR produce several outlier points that vary with corresponding values with the stepwise scheme (x-axis). The
dramatically from the reference results computed with the dashed red lines indicate the region of ±1.0 kcal/mol difference. The
stepwise scheme. With the SSC(2) potential, results are in corresponding standard deviations are plotted as gray error bars and
also shown in the small blue inset plots. The calculated |ϵ|s are
positive definite values and not normally distributed; hence their
standard deviations (shown in parentheses) should not be interpreted
as indicative of data range.
the original AMBER softcore scheme and in Figure 5 for the

proposed SSC(2) scheme. Plots show the comparison between
the stepwise scheme (x-axis) and the reported concerted
scheme (y-axis). The dashed red lines indicate the region of
±1.0 kcal/mol difference. The corresponding standard
deviations are plotted as gray error bars and also shown in
the small blue plots.
The original AMBER softcore scheme results (Figure 4) for
these six targets indicate that the concerted scheme produces
virtually the same RBFE (within 1.0 kcal/mol) results
compared to the stepwise scheme, except one outlier in the
Jnk1 case and one in the MCL1 case (which is out of range on
the plot with a ΔΔG = 13.46 resulting from an unstable
Figure 3. RBFE results for p38 ligands: Upper panels: simulation particle collapse problem shown in Figure S1 in Supporting
results with the original AMBER softcore potential parameter set and Information). The standard deviations from the concerted and
the linear combination of the softcore potentials, analyzed by TI (left the stepwise schemes are also roughly similar and correlated.
panels) and MBAR (right panels). Lower panels: simulation results The proposed SSC(2) scheme results (Figure 5) are
with the proposed second-order smoothstep function SSC(2) and the qualitatively similar, except the outliers of Jnk1 and MCL1
softcore potential parameter set (α = 0.2, β = 50 Å2), analyzed by TI
(left panels) and MBAR (right panels). Plots show the comparison
now are eliminated. Nevertheless, comparing the correlation
between the stepwise scheme (x-axis) and the concerted scheme (y- coefficients (R2), the proposed SSC(2) scheme delivers better
axis). The dashed red lines indicate the region of ±1.0 kcal/mol results for MCL1 (0.80 vs 0.99) and Jnk1 (0.92 vs 0.99). The
difference. The calculated |ϵ|s are positive definite values and not same trend can be found in the regression errors as well in
normally distributed; hence their standard deviations (shown in MCL1 (original:0.39 vs SSC(2):0.20 kcal/mol). For the
parentheses) should not be interpreted as indicative of data range. thrombin case, the agreement with the stepwise protocol is
are not resolved by SSC(2). Hence, as a community, it is

important to continue to build up benchmark datasets and
document their known pathologies.
The choice of α/β inevitably impacts the behavior of the
softcore potentials significantly. As already demonstrated and
discussed in the Results section, the proposed SSC(2) scheme
does not necessary make the corresponding softcore function
better for every set of α/β. Instead, the proposed SSC(2)
scheme eliminates the endpoint catastrophe and mitigates the
large gradient-jump problem and hence provides a more stable
platform for selecting the best α/β to reduce the particle
collapse problems and other possible irregularities of the
softcore function potentials. In the present work, we
demonstrated that α = 0.2 and β = 50 Å2 perform reasonably
well for all cases we tested. Nevertheless, this parameter set is
not perfect for all cases; for example, as in the case of Na+/0,
Figure 5. Results (y-axis) of RBFE for six targets of the Wang et al. the ⟨∂U/∂λ⟩λ curve is still not very smooth at λ ≈ 0.5 because
dataset using the concerted scheme and the SSC(2) potential (α = of minor effect of particle collapse. In fact, in some cases of the
0.2, β = 50 Å2) are compared with corresponding values with the
stepwise scheme (x-axis). The dashed red lines indicate the region of
Tyk2 and MCL1 datasets, the proposed approach yields larger
±1.0 kcal/mol difference. The corresponding standard deviations are standard deviations compared to the original protocol (see
plotted as gray error bars and also shown in the small blue inset plots. Section 3 of the Supporting Information). This suggests that
The calculated |ϵ|s are positive definite values and not normally further exploration of the form of the softcore potential and
distributed; hence their standard deviations (shown in parentheses) optimization of parameters therein may lead to even more
should not be interpreted as indicative of data range. stable and robust concerted alchemical pathways. In this sense,
it is the hope that the SSC potential introduced here is a
worse for SSC(2) in terms of R 2 [original:0.94 vs valuable step forward.
SSC(2):0.84]. In both protocols, the regression errors are Other authors have also considered alchemical path
less than 0.15 kcal/mol with only 10 data points. selection in detail. Most notably, this has been done in the
Hence, for these six targets, the proposed SSC(2) scheme context of nonequilibrium schemes where the path implies a
performs at least as well as the original AMBER softcore time-dependent protocol.89,90 Indeed, there has recently been
approach, and they produce virtually the same RBFE (within renewed interest in this approach for ligand-binding calcu-
1.0 kcal/mol) results compared to the stepwise scheme, except lations because of its focus on explicit bound/unbound
few points near or on the 1.0 kcal/mol error lines, which all states.91 However, this technique is not considered here, and
have large errors in both the concerted and stepwise schemes. it is not obvious that the developments here would transfer.
This might imply that other hidden issues besides the softcore This is because the success of nonequilibrium methods seems
potentials need to be further investigated. to be more strongly conditioned on fluctuations at the
■ DISCUSSION
Although it is one of the most pivotal decisions in setting up
endpoints, both in terms of their magnitude and timescale.92
Although this is essentially the same as the endpoint
catastrophe, the current approach is also concerned with
free energy simulations, the choice of alchemical pathway discontinuities along the path, as these are more central in TI.
essentially remains an unsolved issue. For many trans- Similar in spirit, but slightly different in details, is the closely
formations, existing softcore potential implementations are related work first introduced by Hritz and Oostenbrink93 in the
stable and lead to numerically satisfactory results. However, context of replica exchange molecular dynamics using softcore
many seemingly benign transformations can sporadically and/ potentials and described in further detail by Riniker and co-
or unexpectedly become unstable with no obvious indication workers94 where the use of third-order polynomials enables
as to the underlying issue. To be clear, this is a separate different λ-dependency (referred to a “individual Lambdas”)
problem from the standard simulation issues of adequate for calculation of relative free energies. The form of the
statistical sampling and accurate force field modeling. Indeed, softcore potential also has been shown to have impact on the
the issue is considerably more fundamental because a ability to predict λ derivatives at nonsimulated points in
numerically unstable algorithm can neither sample exhaustively extended TI methods.95 Shirts and co-workers have considered
nor correctly characterize even a “perfect” physical model. The the problem of finding a minimum variance path in equilibrium
previous sections showed several examples where these failures alchemical schemes and found that the fluctuations at
occur using the standard softcore implementation in AMBER. intermediate states (especially, the timescales thereof) are of
Although we do not present specific evidence here, we feel that critical importance.44,45 It has been proposed to overcome the
the data provides a reasonable indication that other related endpoint catastrophe through a capped form of softcore
implementations would likely lead to similar failures on at least potential by Buelens and Grubmü l ler 41 and Pal and
some of these examples. Regardless, the SSC(2) potential Gallicchio.42 Recently, Pal and Gallicchio proposed a general
resolves the issues in nearly all instances, except when the design strategy for the alchemical pathways.42
overall certainty is precluded by a system that is challenging for Besides the various softcore potential forms and alchemical
other reasons (e.g., slow degrees of freedom or high path selections, there are theoretical and technical issues that
variability). Even so, this is still an absence of evidence that need be taken into consideration, such as the treatment of
SSC(2) fails and not evidence that SSC(2) does not fail nonbonded40 interactions within the softcore region and the
there may indeed be situations where the issues outlined here bonded interactions across the common core/softcore
boundary.43,96,97 Although this topic is beyond the scope of the with less computational effort than the stepwise scheme and is
present work, the solutions to these issues have been recently better suited for enhanced sampling methods with more
implemented and extensively validated and tested in advanced, adaptive λ scheduling requirements.
AMBER18,60 fixing some long-standing problems with
previous versions, along with the presented SSC scheme
developed here and implemented as the default in the
AMBER20 release.61
■
*
ASSOCIATED CONTENT
sı Supporting Information
The stepwise scheme has been widely thought to be the The Supporting Information is available free of charge at
most stable procedure for performing TI calculations. The
https://pubs.acs.org/doi/10.1021/acs.jctc.0c00237.
results here show that employing the SSC potential, TI
calculations can be performed in a single concerted trans- Direct space implementation of the electrostatic
formation without loss of precision compared to the stepwise interaction, simulation setup and protocols, representa-
scheme. Hence, the smoothstep potential should be the choice tive ⟨∂U/∂λ⟩λ curves, plots of the smooth step functions
in situations where concerted simulations are preferred. of different orders and their derivatives, comparison of
The calculations shown in this paper mainly utilized the TI the original softcore and SSC(2) functions with different
method. Nonetheless, our data suggests that many of the values of α and β, and results of smoothstep functions
pathological problems discussed in the context of TI are also with different orders (PDF)
problematic for other free energy approaches such as TP62
with BAR71 or MBAR75 analysis. Our results of PTP1B and
p38 (Figures 2 and 3) demonstrate that the SSC(2) potential
also eliminates the observed problems that occur with TP
■ AUTHOR INFORMATION
Corresponding Author
methods with MBAR analysis and converge to the same result
as TI. Darrin M. York − Laboratory for Biomolecular Simulation
Furthermore, the proposed SSC(2) scheme is well suited for Research, Institute for Quantitative Biomedicine and
advanced λ-scheduling optimization and enhanced sampling Department of Chemistry and Chemical Biology, Rutgers
schemes in the λ-space where a single-pass concerted λ University, Piscataway, New Jersey 08854, United States;
transformation is desirable, including λ dynamics,47−50 orcid.org/0000-0002-9193-7055; Email: Darrin.York@
Hamiltonian replica exchange methods,51−55 adaptive bias- rutgers.edu
ing,48,56,57 or self-adjusted mixture sampling58,59 methods. We
Authors
are actively investigating possible incorporation of the
proposed SSC(2) potential with these techniques. Tai-Sung Lee − Laboratory for Biomolecular Simulation
■
Research, Institute for Quantitative Biomedicine and
CONCLUSIONS Department of Chemistry and Chemical Biology, Rutgers
University, Piscataway, New Jersey 08854, United States;
In conclusion, we propose a novel second-order smooth step orcid.org/0000-0003-2110-2279
softcore potential, SSC(2), to overcome the endpoint Zhixiong Lin − Silicon Therapeutics LLC, Boston, Massachusetts
catastrophe, particle collapse, and large gradient-jump 02111, United States
problems routinely encountered in alchemical free energy Bryce K. Allen − Silicon Therapeutics LLC, Boston,
simulations using concerted transformations. These problems Massachusetts 02111, United States; orcid.org/0000-0002-
stem from poor phase space overlap, imbalance of Coulomb 0804-8127
attraction and exchange repulsion, and thermodynamic Charles Lin − Silicon Therapeutics LLC, Boston, Massachusetts
derivative terms in the softcore potential that lead to sensitivity 02111, United States; orcid.org/0000-0002-5461-6690
to softcore α and particularly β parameters. The SSC(2) Brian K. Radak − Silicon Therapeutics LLC, Boston,
potential with α = 0.2 and β = 50 Å2 has been demonstrated to Massachusetts 02111, United States; orcid.org/0000-0001-
overcome these problems for a broad set of alchemical 8628-5972
transformations used in the calculation of hydration free Yujun Tao − Laboratory for Biomolecular Simulation Research,
energies and RBFEs. The key characteristic of the SSC Institute for Quantitative Biomedicine and Department of
function is that the weights used in the alchemical trans- Chemistry and Chemical Biology, Rutgers University,
formation have derivatives that vanish at the transformation Piscataway, New Jersey 08854, United States; orcid.org/
endpoints (λ = 0 and 1) and enable smooth adjustment of the 0000-0002-4520-941X
λ-dependent terms in the potential. This in turn allows Hsu-Chun Tsai − Laboratory for Biomolecular Simulation
Coulomb attraction and exchange repulsions to be rebalanced Research, Institute for Quantitative Biomedicine and
so as to avoid introduction of artificial minima, where a particle Department of Chemistry and Chemical Biology, Rutgers
in the softcore region can collapse on a neighboring particle at University, Piscataway, New Jersey 08854, United States;
intermediate λ states. Results are examined for edge cases orcid.org/0000-0001-7027-5649
where the original AMBER softcore potential is observed to Woody Sherman − Silicon Therapeutics LLC, Boston,
fail, and the SSC(2) potential is shown to remain accurate. The Massachusetts 02111, United States; orcid.org/0000-0001-
SSC(2) potential is further tested against a broad set of 9079-1376
hydration free energy and RBFEs for the Wang et al. dataset
containing 314 ligand mutations and spanning 8 protein Complete contact information is available at:
targets. The SCC(2) potential developed here is demonstrated https://pubs.acs.org/10.1021/acs.jctc.0c00237
to be highly robust, leading to precise free energy values for all
systems considered here. The SSC(2) potential has the Notes
advantage that it can be used in concerted transformations The authors declare no competing financial interest.
Journal of Chemical Theory and Computation
■
pubs.acs.org/JCTC Article
ACKNOWLEDGMENTS binding free energy calculations within the amber molecular dynamics
package. J. Comput. Chem. 2018, 39, 1354−1358.
The authors are grateful for the financial support provided by (16) Giese, T. J.; York, D. M. A GPU-Accelerated Parameter
the National Institutes of Health (no. GM107485 to DMY). Interpolation Thermodynamic Integration Free Energy Method. J.
Computational resources were provided by the Office of Chem. Theory Comput. 2018, 14, 1564−1582.
Advanced Research Computing (OARC) at Rutgers, The State (17) Lee, T.-S.; Cerutti, D. S.; Mermelstein, D.; Lin, C.; LeGrand, S.;
University of New Jersey, the National Institutes of Health Giese, T. J.; Roitberg, A.; Case, D. A.; Walker, R. C.; York, D. M.
under grant no. S10OD012346, the Blue Waters sustained- GPU-Accelerated Molecular Dynamics and Free Energy Methods in
petascale computing project (NSF OCI 07-25070, PRAC OCI- Amber18: Performance Enhancements and New Features. J. Chem.
1515572), and by the Extreme Science and Engineering Inf. Model. 2018, 58, 2043−2050.
Discovery Environment (XSEDE), which is supported by (18) Song, L. F.; Lee, T.-S.; Zhu, C.; York, D. M.; Merz, K. M., Jr.
National Science Foundation grant number no. ACI- Using AMBER18 for Relative Free Energy Calculations. J. Chem. Inf.
1548562.98 This work used the XSEDE resource COMET Model. 2019, 59, 3128−3135.
and COMET GPU at SDSC through allocation TG- (19) Free Energy Calculations: Theory and Applications in Chemistry
CHE190067. We gratefully acknowledge the support of the and Biology; Springer Series in Chemical Physics; Chipot, C.,
Pohorille, A., Eds.; Springer: New York, 2007; Vol. 86.
NVIDIA Corporation with the donation of several Pascal,
(20) Jorgensen, W. L. Efficient drug lead discovery and optimization.
Volta, and Turing GPUs and the GPU-time of a GPU-cluster Acc. Chem. Res. 2009, 42, 724−733.
where the reported benchmark results were performed.
■
(21) Mobley, D. L.; Dill, K. A. Binding of Small-Molecule Ligands to
Proteins: What You See Is Not Always What You Get. Structure 2009,
REFERENCES 17, 489−498.
(1) Stone, J. E.; Phillips, J. C.; Freddolino, P. L.; Hardy, D. J.; (22) Christ, C. D.; Mark, A. E.; van Gunsteren, W. Basic Ingredients
Trabuco, L. G.; Schulten, K. Accelerating molecular modeling of Free Energy Calculations: A Review. J. Comput. Chem. 2010, 31,
applications with graphics processors. Comput. Chem. 2007, 28, 1569−1582.
2618−2640. (23) Foloppe, N.; Hubbard, R. Towards Predictive Ligand Design
(2) Anderson, J. A.; Lorenz, C. D.; Travesset, A. General purpose With Free-Energy Based Computational Methods? Curr. Med. Chem.
molecular dynamics simulations fully implemented on graphics 2006, 13, 3583−3608.
processing units. J. Comput. Phys. 2008, 227, 5342−5359. (24) Pohorille, A.; Jarzynski, C.; Chipot, C. Good Practices in Free-
(3) Hardy, D. J.; Stone, J. E.; Schulten, K. Multilevel Summation of Energy Calculations. J. Phys. Chem. B 2010, 114, 10235−10253.
Electrostatic Potentials Using Graphics Processing Units. Parallel (25) Michel, J.; Essex, J. W. Prediction of protein-ligand binding
Comput. 2009, 35, 164−177. affinity by free energy simulations: assumptions, pitfalls and
(4) Harvey, M. J.; Giupponi, G.; Fabritiis, G. D. ACEMD: expectations. J. Comput.-Aided Mol. Des. 2010, 24, 639−658.
Accelerating Biomolecular Dynamics in the Microsecond Time (26) Gallicchio, E.; Levy, R. M. Advances in all atom sampling
Scale. J. Chem. Theory Comput. 2009, 5, 1632−1639. methods for modeling protein-ligand binding affinities. Curr. Opin.
(5) Harvey, M. J.; de Fabritiis, G. An Implementation of the Smooth Struct. Biol. 2011, 21, 161−166.
Particle Mesh Ewald Method on GPU Hardware. J. Chem. Theory (27) Chodera, J. D.; Mobley, D. L.; Shirts, M. R.; Dixon, R. W.;
Comput. 2009, 5, 2371−2377. Branson, K.; Pande, V. S. Alchemical free energy methods for drug
(6) Stone, J. E.; Hardy, D. J.; Ufimtsev, I. S.; Schulten, K. GPU- discovery: progress and challenges. Curr. Opin. Struct. Biol. 2011, 21,
accelerated molecular modeling coming of age. J. Mol. Graph. Model. 150−160.
2010, 29, 116−125. (28) Mobley, D. L.; Klimovich, P. V. Perspective: Alchemical free
(7) Farber, R. M. Topical perspective on massive threading and energy calculations for drug discovery. J. Chem. Phys. 2012, 137,
parallelism. J. Mol. Graph. Model. 2011, 30, 82−89. 230901.
(8) Götz, A.; Williamson, M. J.; Xu, D.; Poole, D.; Le Grand, S.; (29) Gumbart, J. C.; Roux, B.; Chipot, C. Standard Binding Free
Walker, R. C. Routine microsecond molecular dynamics simulations Energies from Computer Simulations: What Is the Best Strategy? J.
with AMBER on GPUs. 1. Generalized Born. J. Chem. Theory Comput.
Chem. Theory Comput. 2013, 9, 794−802.
2012, 8, 1542.
(30) Hansen, N.; van Gunsteren, W. F. Practical Aspects of Free-
(9) Eastman, P.; et al. OpenMM 4: A Reusable, Extensible,
Energy Calculations: A Review. J. Chem. Theory Comput. 2014, 10,
Hardware Independent Library for High Performance Molecular
2632−2647.
Simulation. J. Chem. Theory Comput. 2013, 9, 461−469.
(31) Cournia, Z.; Allen, B.; Sherman, W. Relative Binding Free
(10) Salomon-Ferrer, R.; Götz, A. W.; Poole, D.; Le Grand, S.;
Walker, R. C. Routine microsecond molecular dynamics simulations Energy Calculations in Drug Discovery: Recent Advances and
with AMBER on GPUs. 2. Explicit solvent Particle Mesh Ewald. J. Practical Considerations. J. Chem. Inf. Model. 2017, 57, 2911−2937.
Chem. Theory Comput. 2013, 9, 3878−3888. (32) de Ruiter, A.; Oostenbrink, C. Advances in the calculation of
(11) Chipot, C. Frontiers in free-energy calculations of biological binding free energies. Curr. Opin. Struct. Biol. 2020, 61, 207−212.
systems. Wiley Interdiscip. Rev.: Comput. Mol. Sci. 2014, 4, 71−89. (33) Mezei, M. Polynomial path for the calculation of liquid state
(12) Eastman, P.; Swails, J.; Chodera, J. D.; McGibbon, R. T.; Zhao, free energies from computer simulations tested on liquid water. J.
Y.; Beauchamp, K. A.; Wang, L.-P.; Simmonett, A. C.; Harrigan, M. Comput. Chem. 1992, 13, 651−656.
P.; Stern, C. D.; Wiewiora, R. P.; Brooks, B. R.; Pande, V. S. (34) Resat, H.; Mezei, M. Studies on free energy calculations. I.
OpenMM 7: Rapid development of high performance algorithms for Thermodynamic integration using a polynomial path. J. Chem. Phys.
molecular dynamics. PLoS Comput. Biol. 2017, 13, 1005659. 1993, 99, 6052−6061.
(13) Abel, R.; Wang, L.; Harder, E. D.; Berne, B. J.; Friesner, R. A. (35) Simonson, T. Free energy of particle insertion. Mol. Phys. 1993,
Advancing Drug Discovery through Enhanced Free Energy Calcu- 80, 441−447.
lations. Acc. Chem. Res. 2017, 50, 1625−1632. (36) Beutler, T. C.; Mark, A. E.; van Schaik, R. C.; Gerber, P. R.; van
(14) Lee, T.-S.; Hu, Y.; Sherborne, B.; Guo, Z.; York, D. M. Toward Gunsteren, W. F. Avoiding singularities and numerical instabilities in
Fast and Accurate Binding Affinity Prediction with pmemdGTI: An free energy calculations based on molecular simulations. Chem. Phys.
Efficient Implementation of GPU-Accelerated Thermodynamic Lett. 1994, 222, 529−539.
Integration. J. Chem. Theory Comput. 2017, 13, 3077−3084. (37) Steinbrecher, T.; Mobley, D. L.; Case, D. A. Nonlinear scaling
(15) Mermelstein, D. J.; Lin, C.; Nelson, G.; Kretsch, R.; schemes for Lennard-Jones interactions in free energy calculations. J.
McCammon, J. A.; Walker, R. C. Fast and flexible GPU accelerated Chem. Phys. 2007, 127, 214108.
(38) Steinbrecher, T.; Joung, I.; Case, D. A. Soft-Core Potentials in (58) Carlson, D. E.; Stinson, P.; Pakman, A.; Paninski, L. Proceedings
Thermodynamic Integration: Comparing One- and Two-Step Trans- of the 33rd Interna-Tional Conference on Machine Learning; ICML’16;
formations. J. Comput. Chem. 2011, 32, 3253−3263. JMLR.org: New York, NY, USA, 2016; Vol. 48; pp 2896−2905.
(39) Gapsys, V.; Seeliger, D.; de Groot, B. L. New Soft-Core (59) Tan, Z. Optimally Adjusted Mixture Sampling and Locally
Potential Function for Molecular Dynamics Based Alchemical Free Weighted Histogram Analysis. J. Comput. Graph Stat. 2017, 26, 54−
Energy Calculations. J. Chem. Theory Comput. 2012, 8, 2373−2382. 65.
(40) Gapsys, V.; Khabiri, M.; de Groot, B. L.; Freddolino, P. L. (60) Tsai, H.-C.; Tao, Y.; Lee, T.-S.; Merz, K. M., Jr.; York, D. M.
Comment on “Deficiencies in Molecular Dynamics Simulation-Based Validation of Free Energy Methods in AMBER. J. Chem. Inf. Model.
Prediction of Protein-DNA Binding Free Energy Landscapes”. J. Phys. 2020, DOI: 10.1021/acs.jcim.0c00285.
Chem. B 2020, 124, 1115−1123. (61) Case, D. A. et al. AMBER 20; University of California: San
(41) Buelens, F. P.; Grubmüller, H. Linear-scaling soft-core scheme Francisco: San Francisco, CA, 2020.
for alchemical free energy calculations. J. Comput. Chem. 2012, 33, (62) Zwanzig, R. W. High-temperature equation of state by a
25−33. perturbation method. I. Nonpolar gases. J. Chem. Phys. 1954, 22,
(42) Pal, R. K.; Gallicchio, E. Perturbation potentials to overcome 1420−1426.
order/disorder transitions in alchemical binding free energy (63) Kirkwood, J. G. Statistical mechanics of fluid mixtures. J. Chem.
calculations. J. Chem. Phys. 2019, 151, 124116. Phys. 1935, 3, 300−313.
(43) Jiang, W.; Chipot, C.; Roux, B. Computing Relative Binding (64) Straatsma, T. P.; Berendsen, H. J. C. Free energy of ionic
Affinity of Ligands to Receptor: An Effective Hybrid Single-Dual- hydration: Analysis of a thermodynamic integration technique to
Topology Free-Energy Perturbation Approach in NAMD. J. Chem. Inf. evaluate free energy differences by molecular dynamics simulations. J.
Model. 2019, 59, 3794−3802. Chem. Phys. 1988, 89, 5876−5886.
(44) Pham, T. T.; Shirts, M. R. Identifying low variance pathways for (65) Jarzynski, C. Nonequilibrium equality for free energy
free energy calculations of molecular transformations in solution differences. Phys. Rev. Lett. 1997, 78, 2690−2693.
phase. J. Chem. Phys. 2011, 135, 034114. (66) Crooks, G. E. Path-ensemble averages in systems driven far
(45) Pham, T. T.; Shirts, M. R. Optimal pairwise and non-pairwise from equilibrium. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat.
alchemical pathways for free energy calculations of molecular Interdiscip. Top. 2000, 61, 2361−2366.
transformation in solution phase. J. Chem. Phys. 2012, 136, 124120. (67) Boresch, S.; Woodcock, H. L. Convergence of single-step free
(46) Klimovich, P. V.; Shirts, M. R.; Mobley, D. L. Guidelines for the energy perturbation. Mol. Phys. 2017, 115, 1200−1213.
analysis of free energy calculations. J. Comput.-Aided Mol. Des. 2015, (68) Gapsys, V.; Michielssens, S.; Peters, J. H.; de Groot, B. L.;
29, 397−411. Leonov, H. Calculation of binding free energies. Methods Mol. Biol.
(47) Ding, X.; Vilseck, J. Z.; Hayes, R. L.; Brooks, C. L. Gibbs 2015, 1215, 173−209.
Sampler-Based λ-Dynamics and Rao-Blackwell Estimator for (69) Jeong, D.; Andricioaei, I. Reconstructing equilibrium entropy
Alchemical Free Energy Calculation. J. Chem. Theory Comput. 2017, and enthalpy profiles from non-equilibrium pulling. J. Chem. Phys.
13, 2501−2510. 2013, 138, 114110.
(48) Hayes, R. L.; Armacost, K. A.; Vilseck, J. Z.; Brooks, C. L. (70) Wei, D.; Song, Y.; Wang, F. A simple molecular mechanics
Adaptive Landscape Flattening Accelerates Sampling of Alchemical potential for μm scale graphene simulations from the adaptive force
Space in Multisite λ Dynamics. J. Phys. Chem. B 2017, 121, 3626− matching method. J. Chem. Phys. 2011, 134, 184704.
3635. (71) Bennett, C. H. Efficient estimation of free energy differences
(49) Guo, Z.; Brooks, C. L. Rapid Screening of Binding Affinities: from Monte Carlo data. J. Comput. Phys. 1976, 22, 245−268.
Application of the λ- Dynamics Method to a Trypsin-Inhibitor (72) Torrie, G. M.; Valleau, J. P. Nonphysical sampling distributions
System. J. Am. Chem. Soc. 1998, 120, 1920−1921. in Monte Carlo freeenergy estimation: Umbrella sampling. J. Comput.
(50) Guo, Z.; Brooks, C. L., III; Kong, X. Efficient and Flexible Phys. 1977, 23, 187−199.
Algorithm for Free Energy Calculations Using the λ-Dynamics (73) Lu, N.; Kofke, D. A. Accuracy of free-energy perturbation
Approach. J. Phys. Chem. B 1998, 102, 2032−2036. calculations in molecular simulation. I. Modeling. J. Chem. Phys. 2001,
(51) Jiang, W.; Roux, B. Free energy perturbation Hamiltonian 114, 7303−7311.
replica-exchange molecular dynamics (FEP/H-REMD) for absolute (74) Shirts, M. R.; Pande, V. S. Comparison of efficiency and bias of
ligand binding free energy calculations. J. Chem. Theory Comput. 2010, free energies computed by exponential averaging, the Bennett
6, 2559−2565. acceptance ratio, and thermodynamic integration. J. Chem. Phys.
(52) Arrar, M.; de Oliveira, C. A. F.; Fajer, M.; Sinko, W.; 2005, 122, 144107.
McCammon, J. A. w-REXAMD: A Hamiltonian replica exchange (75) Shirts, M. R.; Chodera, J. D. Statistically optimal analysis of
approach to improve free energy calculations for systems with samples from multiple equilibrium states. J. Chem. Phys. 2008, 129,
kinetically trapped conformations. J. Chem. Theory Comput. 2013, 9, 124105.
18−23. (76) Darden, T.; York, D.; Pedersen, L. Particle mesh Ewald: An N
(53) Itoh, S. G.; Okumura, H. Hamiltonian replica-permutation log(N) method for Ewald sums in large systems. J. Chem. Phys. 1993,
method and its applications to an alanine dipeptide and amyloid- 98, 10089−10092.
β(29-42) peptides. J. Comput. Chem. 2013, 34, 2493−2497. (77) in’t Veld, P. j.; Ismail, A. E.; Grest, G. S. Application of Ewald
(54) Armacost, K. A.; Goh, G. B.; Brooks, C. L. Biasing Potential summations to long-range dispersion forces. J. Chem. Phys. 2007, 127,
Replica Exchange Multisite λ-Dynamics for Efficient Free Energy 144711.
Calculations. J. Chem. Theory Comput. 2015, 11, 1267−1277. (78) Le Grand, S.; Götz, A. W.; Walker, R. C. SPFP: Speed without
(55) Yang, M.; Huang, J.; MacKerell, A. D., Jr. Enhanced compromise−A mixed precision model for GPU accelerated
Conformational Sampling Using Replica Exchange with Concurrent molecular dynamics simulations. Comput. Phys. Commun. 2013, 184,
Solute Scaling and Hamiltonian Biasing Realized in One Dimension. 374−380.
J. Chem. Theory Comput. 2015, 11, 2855−2867. (79) Maier, J. A.; Martinez, C.; Kasavajhala, K.; Wickstrom, L.;
(56) Babin, V.; Roland, C.; Sagui, C. Adaptively biased molecular Hauser, K. E.; Simmerling, C. ff14SB: Improving the Accuracy of
dynamics for free energy calculations. J. Chem. Phys. 2008, 128, Protein Side Chain and Backbone Parameters from ff99SB. J. Chem.
134101. Theory Comput. 2015, 11, 3696−3713.
(57) Darve, E.; Rodríguez-Gómez, D.; Pohorille, A. Adaptive biasing (80) Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.; Case, D.
force method for scalar and vector free energy calculations. J. Chem. A. Development and testing of a general amber force field. J. Comput.
Phys. 2008, 128, 144120. Chem. 2004, 25, 1157−1174.
(81) Jakalian, A.; Jack, D. B.; Bayly, C. I. Fast, efficient generation of

high-quality atomic charges. AM1-BCC model: II. parameterization
and validation. J. Comput. Chem. 2002, 23, 1623−1641.
(82) Jakalian, A.; Bush, B. L.; Jack, D. B.; Bayly, C. I. Fast, efficient
generation of high-quality atomic charges. AM1-BCC model: I.
method. J. Comput. Chem. 2000, 21, 132−146.
(83) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R.
W.; Klein, M. L. Comparison of simple potential functions for
simulating liquid water. J. Chem. Phys. 1983, 79, 926−935.
(84) Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C. Numerical
Integration of the Cartesian Equations of Motion of a System with
Constraints: Molecular Dynamics of n-Alkanes. J. Comput. Phys. 1977,
23, 327−341.
(85) Miyamoto, S.; Kollman, P. A. SETTLE: An analytic version of
the SHAKE and RATTLE algorithms for rigid water models. J.
Comput. Chem. 1992, 13, 952−962.
(86) Lee, Y.-K.; Parks, D. J.; Lu, T.; Thieu, T. V.; Markotan, T.; Pan,
W.; McComsey, D. F.; Milkiewicz, K. L.; Crysler, C. S.; Ninan, N.;
Abad, M. C.; Giardino, E. C.; Maryanoff, B. E.; Damiano, B. P.;
Player, M. R. 7-fluoroindazoles as potent and selective inhibitors of
factor Xa. J. Med. Chem. 2008, 51, 282−297.
(87) Loeffler, H. H.; Bosisio, S.; Duarte Ramos Matos, G.; Suh, D.;
Roux, B.; Mobley, D. L.; Michel, J. Reproducibility of Free Energy
Calculations across Different Molecular Simulation Software Pack-
ages. J. Chem. Theory Comput. 2018, 14, 5567−5582.
(88) Wang, L.; et al. Accurate and reliable prediction of relative
ligand binding potency in prospective drug discovery by way of a
modern free-energy calculation protocol and force field. J. Am. Chem.
Soc. 2015, 137, 2695−2703.
(89) Suh, D.; Radak, B. K.; Chipot, C.; Roux, B. Enhanced
configurational sampling with hybrid non-equilibrium molecular
dynamics-Monte Carlo propagator. J. Chem. Phys. 2018, 148, 014101.
(90) Chen, Y.; Roux, B. Enhanced Sampling of an Atomic Model
with Hybrid Nonequilibrium Molecular Dynamics-Monte Carlo
Simulations Guided by a Coarse-Grained Model. J. Chem. Theory
Comput. 2015, 11, 3572−3583.
(91) Gapsys, V.; Pérez-Benito, L.; Aldeghi, M.; Seeliger, D.; van
Vlijmen, H.; Tresadern, G.; de Groot, B. L. Large scale relative
protein ligand binding affinities using nonequilibrium alchemy. Chem.
Sci. 2020, 11, 1140−1152.
(92) Radak, B. K.; Roux, B. Efficiency in nonequilibrium molecular
dynamics Monte Carlo simulations. J. Chem. Phys. 2016, 145, 134109.
(93) Hritz, J.; Oostenbrink, C. Hamiltonian replica exchange
molecular dynamics using softcore interactions. J. Chem. Phys. 2008,
128, 144121.
(94) Riniker, S.; Christ, C. D.; Hansen, H. S.; Hünenberger, P. H.;
Oostenbrink, C.; Steiner, D.; van Gunsteren, W. F. Calculation of
relative free energies for ligand-protein binding, solvation, and
conformational transitions using the GROMOS software. J. Phys.
Chem. B 2011, 115, 13570−13577.
(95) Ruiter, A. d.; Oostenbrink, C. Extended Thermodynamic
Integration: Efficient Prediction of Lambda Derivatives at Non-
simulated Points. J. Chem. Theory Comput. 2016, 12, 4476−4486.
(96) Boresch, S.; Karplus, M. The Role of Bonded Terms in Free
Energy Simulations: 1. Theoretical Analysis. J. Phys. Chem. A 1999,
103, 103−118.
(97) Boresch, S.; Karplus, M. The Role of Bonded Terms in Free
Energy Simulations: 2. Calculation of Their Influence on Free Energy
Differences of Solvation. J. Phys. Chem. A 1999, 103, 119−136.
(98) Towns, J.; Cockerill, T.; Dahan, M.; Foster, I.; Gaither, K.;
Grimshaw, A.; Hazlewood, V.; Lathrop, S.; Lifka, D.; Peterson, G. D.;
Roskies, R.; Scott, J. R.; Wilkins-Diehr, N. XSEDE: Accelerating
Scientific Discovery. Comput. Sci. Eng. 2014, 16, 62−74.

Fep 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fep 2

Uploaded by

Copyright:

Available Formats

pubs.acs.

Improved Alchemical Free Energy Calculations with Optimized

ACCESS Metrics & More Article Recommendations *

ABSTRACT: Progress in the development of GPU-accelerated free energy simulation

© 2020 American Chemical Society https://dx.doi.org/10.1021/acs.jctc.0c00237

ULJ(rij) = 4ϵijÅÅÅÅjjj zzz − jjjj zzzz ÑÑÑÑ

= −4jjjj zzzzÅÅÅÅ12jjjj zzzz − 6jjjj zzzz ÑÑÑÑ

j rij zÅÅ j rij z j rij z ÑÑ

ÅÅÇ dx k ÑÑÑÖx = 0 ÅÅÅÇ dx k ÑÑÑÖx = 1

close agreement (within statistical error estimates, typically less

Figure 2. RBFE results for PTP1B ligands: Upper panels: simulation

statistical error estimates, and the TI and MBAR results are

the original AMBER softcore scheme and in Figure 5 for the

are not resolved by SSC(2). Hence, as a community, it is

(81) Jakalian, A.; Jack, D. B.; Bayly, C. I. Fast, efficient generation of

You might also like