OPLS3e: Extending Force Field Coverage for Drug-Like Small

Subscriber access provided by UNIV OF NEW ENGLAND ARMIDALE

Molecular Mechanics

OPLS3e: Extending Force Field Coverage for Drug-Like Small Molecules Katarina Roos, Chuanjie Wu, Wolfgang Damm, Mark Reboul, James M. Stevenson, Chao Lu, Markus K. Dahlgren, Sayan Mondal, Wei Chen, Lingle Wang, Robert Abel, Richard A Friesner, and Edward D. Harder J. Chem. Theory Comput., Just Accepted Manuscript • DOI: 10.1021/acs.jctc.8b01026 • Publication Date (Web): 15 Feb 2019 Downloaded from http://pubs.acs.org on February 15, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

OPLS3e: Extending Force Field Coverage for DrugLike Small Molecules Katarina Roos2,3, Chuanjie Wu1, Wolfgang Damm1, Mark Reboul1, James M. Stevenson1, Chao Lu1, Markus K. Dahlgren1, Sayan Mondal1, Wei Chen1, Lingle Wang1, Robert Abel1, Richard A. Friesner2, Edward D. Harder*1 1Schrodinger,

Inc., 120 West 45th Street, New York, New York

10036, United States,

2Department

of Chemistry, Columbia

University, 3000 Broadway, New York, New York 10027, United States, 3Department of Cell and Molecular Biology, Uppsala University, Biomedical Centre, BOX 596, SE-751 24 Uppsala, Sweden

ABSTRACT

Building upon the OPLS3 force field we report on an enhanced model, OPLS3e, that further extends its coverage of medicinally relevant chemical

space

transferability.

by

addressing

limitations

in

chemotype

OPLS3e accomplishes this by incorporating new

ACS Paragon Plus Environment

1

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 53

parameter types that recognize moieties with greater chemical specificity

and

integrating

an

on-the-fly

approach to the assignment of partial charges.

parameterization As a consequence,

OPLS3e leads to greater accuracy against performance benchmarks that assess small molecule conformational propensities, solvation and protein-ligand binding.

I. Introduction Free

energy

approaches

that

utilize

all-atom

molecular

dynamics simulations are emerging as a powerful tool in the design of new molecules and materials. Spurring interest in these methods is the promise of a high degree of predictive accuracy owing to the

physical

rigor

with

which

these

methods

account

for the

composite atomistic and configurational complexity of molecular systems. ability to

In principle, their accuracy is only limited by the achieve

adequate

phase

space

sampling

and

by the

veracity of the model potential. Because of their speed, the class of approximate molecular models commonly referred to as force fields offer a tractable means for applying free energy methods to the study of phenomena in complex biological and materials systems. A notable example is the application of free energy perturbation theory (FEP) to calculate ligand binding potencies in structurebased drug discovery projects.1-5

The adoption of this technology

in

traction

active

projects

is

gaining


as

advances

to

the

2

Page 3 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


technology result in greater automation, delivering more robust results

with

greater

expediency.6-11

As

the

capacity

of

the

technology continues to improve and sees greater usage on a wider regime of systems it is imperative that the development of the force field continues apace to ensure meaningful and predictive results. Force fields aim to strike a balance between computational expediency and physical salience and accomplish this goal by employing simple potential functions, such as, harmonic terms to represent bond vibrations and fixed-point charges to approximate electrostatic interactions.12-13 These terms depend explicitly upon sets

of

empirically

determined

pertinent physical properties.

parameters

tuned

to

reproduce

Target parameterization data is

typically comprised of quantum mechanical and experimental data for representative and diverse small organic molecules. effective,

a

force

field

needs

to

provide

an

To be

accurate

parameterization of all components of the system that can impact a property of interest. For ligand-protein binding, a particularly noteworthy challenge is the development of a force field that achieves a uniformly high level of accuracy (on the order of 1 kcal/mol) over the many diverse chemistries found within drug-like molecules. Early efforts to build force fields for drug-like small molecules have tended to focus on parameter generalization.14-17 That

is,

automating

the

process

by

which


a

set

of

existing

3


Page 4 of 53

parameters can be translated into assignments for any arbitrary ligand.

Less effort has been placed on building the extensive

basis of parameters needed to cover drug-like chemical space at a high

level

of

accuracy.

The

recent

development

of

OPLS318

represents perhaps the largest concerted parameterization effort aimed at traversing this space and it forms the starting point for the work discussed herein.

To review, notable features of the

OPLS3 model include: 1. An improved protein force field. 2. Extensive

parameterization

of

valence

and

torsional

terms found in drug-like molecules supplemented by an automated version of the fitting protocol (FFBuilder) used to fill coverage gaps. 3. Improvements to the charge model including virtual (offatom centered) sites to better represent lone pair and sigma hole charge distributions. These advances translated to significant improvements in proteinligand binding accuracy using FEP.

Relative to OPLS2005, OPLS3

resulted in a 30% improvement in aggregate RMS error.

Despite

these gains, the current prediction error continues to exceed the expected

experimental

indicating

additional

uncertainties room

for

by

a

improvement

significant in

the

margin,

technology

remains.


4

Page 5 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


The aim of the work presented herein is to address additional outstanding

sources

of

force

field

error

significantly to the residual FEP error.

that

contribute

In particular, we’ve

found that sub-optimal parameter transferability across shared chemotypes

impacting

parameterizations

can

both

the

torsion

significantly

impact

and FEP

partial

charge

accuracy.

To

address this, a new version of the force field, OPLS3e, has been developed. 

Notable improvements include:

A more accurate and transferable torsional parameterization via an extensive optimization of torsion types.



Integration of a ligand-specific, on-the-fly partial charge assignment approach.

The paper is organized as follows.

In section 2, we present

two representative cases that illustrate the source of force field error resulting in significant binding free energy mispredictions. Section 3 discusses the OPLS3e development.

Part 3.A provides a

brief overview, Part 3.B presents the new torsion model including the process of refining torsion types, the extended data sets that were

required

to

train

the

model

and

the

performance

on

conformational energy benchmarks. Part 3.C presents the new charge assignment method and also includes a discussion of the simple condensed phase data sets used to develop and validate the model. In Section 4 we discuss the protein-ligand binding validation. This data set includes a significant extension from previous


5


reports,1,

18

Page 6 of 53

with new ligand and receptor data collected with the

expressed aim of presenting challenging chemotype transferability cases for the force field.

Section 5 then summarizes the results

and discusses future directions.

II.

OPLS3 Force Field Outliers Suboptimal

assignment

of

torsion

and

partial

parameters can lead to erroneous free energy calculations.

charge Panel

A of Figure 1 shows an example of an FEP+ misprediction that arose in a drug discovery program that was due to sub-optimal torsion parameters (only details that directly pertain to the outlier investigation are presented). of

an amide substituent

The ligands differ by a replacement

with an ester.

Experimentally, the

relative binding free energy of the ligands is approximately zero. However,

the

OPLS3-based

FEP+

prediction

significantly

underestimates the amide-pyrrole binding free energy relative to the ester.

An indicator of the underlying model issue driving

the erroneous prediction is shown in Panel B.

The top fragment

represents the substructure of the amide-pyrrole ligand in the FEP calculation. The plot contrasts the associated QM and MM potential energy

scanning

the

dihedral

angle

connecting

the

amide

substituent to the pyrrole ring defined by the NCcn atoms (aromatic atoms in lower case).

The predominant rotamer state in solution

corresponds to an NCcn angle ~ 0.

The predominant state in the


6

Page 7 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


complex corresponds to an NCcn angle ~ 180 which brings the NH of the amide and pyrrole into close proximity facilitating a bidentate contact with a nearby histidine.

As such, to adopt the bioactive

conformation,

derivative

the

amide-pyrrole

degree of conformational work.

needs

to

pay

some

Based on the potential energy

comparison to QM, OPLS3 appears to significantly overestimate this contribution affinity.

consistent

with

the

observed

To verify this hypothesis,

under-predicted

we re-ran the

binding

experiment with a slight modification to OPLS3 (denoted OPLS3*) where the amide-pyrrole torsion of 2-amide-3 methyl pyrrole is refit to match QM. The associated relative amide-to-ester free energy of

OPLS3*

is

-0.23

kcal/mol,

in

excellent

agreement

with

experiment. To explain why OPLS3 errs in matching QM for this dihedral, the training set molecule used to fit the OPLS3 amide-pyrrole dihedral, is shown in panel B of Figure 1 (bottom fragment). Though chemically similar to the ligand substructure of interest, the training set molecule does contain two crucial differences that

ultimately

result

in

poor

quantitative

transferability to the moiety of interest.

parameter

The pyrrole nitrogen

is capped by a methyl group instead of hydrogen and the training set molecule lacks an additional ortho-methyl group.

Because both

features lie one bond away from the atoms used to define the OPLS3 torsion, the same parameters are assigned to both motifs.


The

7


Page 8 of 53

modified version of the force field, OPLS3*, which was trained to the 2-amide-3-pyrrole fragment, shows a similar level of error when

applied

to

the

OPLS3

training

set

molecule,

further

illustrating that a common set of parameters can’t reconcile the energetics

of

both

moieties.

To

resolve

this

incompatibility

issue, the assignment algorithm would need to distinguish between these motifs by taking into account an extended chemical regime that includes these ortho positioned features. Additional

examples

emerged

following

a

similar

theme:

differences in chemistry unaccounted for by the assigned torsion parameter type lead to sub-optimal energetic transferability to a moiety of interest resulting in an FEP+ mis-prediction.

However,

the chemical features driving these errors varied from case to case.

In an effort to identify and minimize instances like this,

in future applications, we set out to systematically refine the chemical basis that defines torsion parameter types to better delineate groups that necessitate distinct parameter assignments. In section 3 we discuss that effort. Our second example illustrates an FEP+ misprediction due to sub-optimal

assignment

of

partial

charges.

We

observed

poor

predictive performance using OPLS3 over a set of inhibitors with cyclic amidine and guanidine cores binding to the two catalytic aspartates of Bace-1.19,

20

As shown in Figure 2, the predicted

binding free energy for replacing an aromatic carbon for a nitrogen


8

Page 9 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(aza derivatives) in the amino-isoindole core is underestimated by 3-4 kcal/mol.

After reviewing the parameters assigned to the

amidine motif we noticed that the partial charges of the aminoaza-isoindole cores differ significantly and consistently from those of the amino-isoindole cores, especially for the amidinium atoms HNC marked in the Figure. We validated this error hypothesis by

fitting

QM charges to

the

ligands

directly

“OPLS3+QMcharges” method outlined in Section 3C.

following the The fitted

partial charges of the amidine motif are in qualitative agreement with the OPLS3 charges for the amino-isoindole core, and consistent for the aza derivatives.

The FEP+ predicted free energy using

OPLS3 with these QM fitted charges leads to good agreement with experiment (Figure 2).20 The reason for this OPLS3 error follows from dependencies of the method on the composition of the training set.

OPLS3 employs

a CM1A-BCC based charge model that combines Cramer-Truhlar CM1A charges21 with specifically fit bond charge correction terms (BCC). The

BCC

corrections

are

primarily

fit

to

QM

electrostatic

potentials (ESP) (using HF/6-31G*) but also include additional refinements against experimental solvation free energies22.

These

BCCs in turn depend on the molecule composition of this training set. The OPLS3 training set includes the bicyclic core of compound 11.

However, the associated aza derivatives are not part of the

OPLS3 training set and the BCCs assigned to the cores in compound


9


Page 10 of 53

13 and 16 instead fall back to parameters based on the next best available

match.

In

this

case,

the

next

best

match

is

a

combination of pyridine and acyclic amidine which transfers poorly to the fused ring substructure of interest. The example illustrates a general need for better equipping the charge model when it encounters ligand substructures that lie outside the force field training set.

To address this, we develop

a ligand-level charge model that assigns partial charges for each new compound based on a combination of fitting to QM derived ESP and restraints to OPLS3 and describe a framework that can easily be further updated in the future.

III.

OPLS3e Developments

A. Overview The functional form of the force field and the development of bond stretching, angle bending and van der Waals parameters is discussed

in

the

OPLS3

unchanged in OPLS3e.

reporting

publication18

and

remains

Changes with respect to OPLS3 include

revisions to the small molecule torsional and charge potentials covered in parts B and C, below. B.

Refined Torsion Types

Torsional

parameters

are

needed

to

describe

the

quantum

mechanical effects associated with stereoelectronic effects and resonance structures.

They play a critical role in determining


10

Page 11 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


molecule

conformational

protein-ligand

binding

energies by

which

modulating

the

sensitively

impacts

conformational

work

needed for a ligand to adopt its bioactive conformation.23

In

OPLS3e, this component of the potential is represented by the same truncated

Fourier

series

described

previously

(see

eq

1

in

reference 18) where the associated V1-4 parameters are determined for each torsion type by fitting to model molecule quantum chemical torsion energy profiles. The profiles were generated as follows: 1. Force field-based torsion scans (in 30 degree increments) are developed from low energy

conformers

and

used

as

the

starting

geometries

in

a

subsequent QM optimization. 2. The quantum chemical profile is resolved via restrained optimization of the molecules at the B3LYP/6-31G* level for each discretized value of the dihedral angle. 3.

Single point M06-2X/cc-pVTZ(-f) calculations are used

to resolve the reference energy surface.

This represents a change

in the QM level used relative to earlier work, which was motivated by

a benchmark comparison

of various

CCSD(T)/CBS, summarized in Table S1.

QM methods relative to

This level is used in the

fitting of all OPLS3e small molecule torsions and is QM reference for all subsequent force field comparisons reported in the paper. The most significant change from OPLS3 is in the definition of the

constituent

torsion

types.

Motivated

by

erroneous

calculations similar to the example discussed in Section 2, we set


11


out

to further optimize

the type definitions

Page 12 of 53

to improve the

recognition of chemical environments that should be treated as distinct.

Our approach can be summarized as follows:

1. for each OPLS3 dihedral of interest, identify additional model molecules that share the same torsion types but differ in their chemical composition; 2. develop associated quantum chemical torsion profiles for the set of molecules; 3. test parameter transferability by training to one member of the molecule set and assessing the accuracy relative to QM for the other molecules; 4. identify the distinguishing chemical features that lead to a significant loss in accuracy; 5. revise the torsion type definitions to encapsulate the new chemical features; and 6. refit the model with the new basis of torsion types. An example illustrating this process is summarized in Table 1. Here, the dihedral of interest is the amino rotation of dimethyl aniline.

For OPLS3, the training set molecule is the otherwise

unsubstituted phenyl derivative.

The test set includes para

substituted methyl, fluoro, chloro, amino, hydroxyl, enolate, nitrile and phenyl.

The parameter transferability quality is

indicated by the root mean square error with respect to QM, over the rotamer energy surface, provided for each fragment.


Good

12

Page 13 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


transferability

is

seen

for

para-methyl,

chloro

and

phenyl

derivatives, where the substituents have little impact on the QM barrier to rotation.

The π-electron fluoro, amino, hydroxyl and

enolate donors significantly weaken the QM barrier to rotation. π-electron

acceptor,

significantly strengthens the QM barrier to rotation.

The trend

The

nitrile

derivative,

which

is

a

is consistent with previous spectroscopic studies.24 The impact these substituents have on the conjugation properties of the torsion in question is not captured by the force field leading to a significant drop in accuracy for these groups. this

limitation,

OPLS3e

encapsulates

this

To address

information

by

delineating its parameter assignment based on the presence of the particular

para

transferability.

substituents

that

significantly

impact

As a consequence, OPLS3e achieves a RMS error

no greater than 0.3 kcal/mol for this data set.

To the 11,845

fragment molecule torsional profiles that constitute the OPLS3 training set18 we added an additional 11,161 fragment profiles to conduct similar parameter transferability tests. In

total,

the

parameter

type

refinements

made

in

OPLS3e

encompass variations along 4 main chemical classes: heteroatom composition,

ortho

substituents,

exocyclic

and

meta/para

substituents.

Figure 3, shows a representative pair of molecules

for each class, with their associated rotamer energies.

Where

in OPLS3, the indicated dihedral parameters are shared for each


13


Page 14 of 53

pair, OPLS3e now treats each moiety as distinct. One consequence of refining the definition of torsion types to be more chemotype specific is that it impacts to what extent chemical substructures will be fully covered by pre-defined well parameterized OPLS3e torsion types.

In Harder et al. we estimated that the fraction

of torsional bonds found in drug-like molecules that would match existing trained OPLS3 torsion types was 93%.

At this stage of

the OPLS3e development that estimate dropped to 68%. To recover a comparable level of statistical coverage an additional 19,558 torsional profiles, for model molecules with missing coverage, were resolved and fit using OPLS3e. That is 30,719 more torsional rotamer sets than what we used for developing OPLS3.

The

concomitant refinements resulted in a total of 146,669 torsion types compared to the 48,142 types in OPLS3. we

provide

an

automated

version

of

And, as with OPLS3,

our

fitting

protocol

(FFBuilder) that can be used to fill in outstanding gaps in OPLS3e coverage. Table 2 contrasts the rotamer energy RMS error for OPLS2005, OPLS3

and

OPLS3e.

The

comparison

excludes

molecules

whose

constituent torsion types are not presently covered by OPLS3. The evaluation is split by molecules used to train and test OPLS3e.

Though OPLS3 (RMSE~1 kcal/mol) performed significantly

better than its older counterpart, OPLS2005 (RMSE~2 kcal/mol) the


14

Page 15 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


new

OPLS3e

parameterization

performs

significantly

better

(RMSE~0.5 kcal/mol).

C. Ligand-Based On-the-Fly Charge Model The OPLS3e charge model retains the same Coulombic fixed charge functional form utilized by OPLS3.

Additional virtual (off-atom

centered) sites for aromatic ring nitrogens and aryl halogens (excluding fluorine) that were added to OPLS3 are retained in OPLS3e.

The difference with respect to OPLS3 is in the method

used to assign partial charges.

As discussed in Section 2, OPLS3

employs a CM1A-BCC scheme that depends on a set of pre-tabulated bond charge corrections predominantly fit to recapitulate the HF/6-31G* electrostatic potential of 11,845 model compounds with additional general,

refinements

this

model

against

has

solvation

demonstrated

an

free

energies.

acceptable

level

In of

accuracy when applied to molecules that lie outside its training set..6-11,

18

However, as evidenced by the Bace-1 outlier shown in

Section 2, application to untrained heterocycles can prove to be problematic. Two approaches that could be applied to ameliorate this issue are: (1) extend the BCC training set to include a growing list of heterocycles25 or (2) parameterize the partial charges for each new ligand by fitting directly to the target QM data (i.e. HF/6-31G* ESP’s) on-the-fly.

Option (1) is akin to the approach taken in


15


refining the torsion model.

Page 16 of 53

For charges, however, we expect this

route to present a greater challenge owing to the expansive nature of drug-like chemical space and the particular sensitivity of charges to their chemical environment. method

maximally

encapsulates

all

Because the on-the-fly the

pertinent

chemical

information of the constituent ligands, we’ve chosen to focus on this approach.

One potential downside is that the compute time

required to generate the QM ESP can be prohibitive for certain FFbased applications such as virtual screening or small molecule conformational search.

However, for MD based applications, such

as FEP, the requisite assignment time (~10 min) remains small relative to the time required for the production simulations. Though HF/6-31G* based charges, in general, perform well when combined with the OPLS force field, there are a limited set of known chemical moieties where QM based charge fitting does not translate

well

addressed

by

to

the

phase.26

condensed

additional

refinement

available condensed phase data.22

of

In

BCC

OPLS3,

this

parameters

was

against

To retain these refinements in

OPLS3e we use a restrained QM fitting approach, where the reference values

are

based

on

OPLS3

charges

and

the

restraints can vary by chemical substructure.

strength

of

the

In principal, this

would allow the model to hew more closely to this reference for moieties

that

required

additional

freely to QM, otherwise.

refinements

and

adapt

more

The restraint values are developed and


16

Page 17 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


validated against a set of small molecule experimental hydration and transfer free energies. The hydration free energy data set is comprised of the Shivakumar et al.22 set of 240 molecules, as well as

an

additional

178

molecules

from

the

FreeSolv

database27

selected to cover a variation of chemical groups while excluding larger compounds with multiple rotatable bonds.

Applying the same

selection criteria, we further expanded our validation sets to include a sub-set of water-toluene and water-chloroform transfer free energies from references 28-30.

FEP+ solvation free energies

were calculated by annihilation of the compound from solvent. Simulations times were 4ns, charge and van der Waals perturbations were run sequentially and transfer free energies were calculated as the difference between annihilation free energies in water and in chloroform/toluene respectively. As a baseline, we explored a charge model with zero restraints to reference values (OPLS3 using QM charges).

Results for these

solvation free energy benchmarks are shown in Table 3. of

the

hydration

free

energy

benchmark

was

The error

observed

to

be

significantly worse using this model versus standard OPLS3 (1.39 kcal/mol versus 1.02 kcal/mol).

Upon closer inspection, the

regression was concentrated in the following functional groups (see Table 4): amines, formates/esters, nitros and the indicated sulfur derivatives.

Representative fragments for these groups are

shown in Figure S1. Amines present a well-known issue where the QM


17


Page 18 of 53

ESP derived partial charges fail to describe the trend in solvation energies of primary, secondary and tertiary amines.26 As discussed previously,

OPLS3

has

addressed

this

by

further

refining

parameters against additional condensed phase data. After

iterative

empirical

testing

against

condensed

phase

benchmarks we settled on a two-tier restraint scheme, with a relatively large restraint for the functional groups mentioned above

and

a

relatively

small

restraint

for

everything

else.

Including low restraints helps ameliorate the problems that the partial charges can suffer from the underdetermined nature of fitting to ESP’s for ligand sized molecules31 and ensure physically relevant charges (e.g. non-negative hydrogens).

The hybrid model,

OPLS3e, compares well to OPLS3 for hydration free energies and shows a marked improvement in the transfer free energy benchmarks (Table

3).

In

particular,

the

chemical

groups

identified

as

problematic to describe with an unrestricted fit to the QM ESP are well described in OPLS3e (Table 4).

It is important to emphasize

here that this scheme is predicated on a posteriori knowledge of (and the availability of adequate existing parameters for) the set of moieties where QM electrostatic potential fitting leads to poor transferability to the condensed phase.

This list of moieties

will require revision if, with the acquisition of new experimental data, additional chemistries are identified that show a similar lack of transferability.


18

Page 19 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


We need to further address the preparation of the molecular conformation used to fit the partial charges. Because fixed charge models do not explicitly account for intramolecular polarization effects, their charge parameters will depend on the particular conformation(s)

used

when

fitting

to

the

QM

ESP.

This

is

especially pertinent for ligand sized molecules in a FEP+ like context.

Here we employ the putative bioactive conformation used

in the FEP calculation. extended charges)22

conformation but

found

As an alternative, we also attempted an (employed

better

by

OPLS3

performance

to using

calculate the

CM1A

bioactive

conformation. As an initial step before fitting charges, the given input

molecule

geometries

are

pre-minimized

with

a

torsional

constraint, to preserve the initially provided conformation while avoiding strain.

In principal, systems where this fixed charge

approach results in parameters that are incongruous with other pertinent

ligand

conformations

(eg.

in

water

solvent)

would

necessitate an electrostatic treatment that explicitly accounts for these polarization effects.32,33

However, we note that in the

FEP+ systems studied in the present work we did not observe significant errors as a consequence of the present approximation.

IV. Application to Protein-Ligand Binding In reference 18, we used a large data set of protein-ligand binding FEP calculations to assess the impact of improvements made


19


to the force field.

Page 20 of 53

Comparing OPLS3 to earlier variants of OPLS

we observed a significant systematic reduction in FEP RMS error with improvements to the quality of the force field.

Here we aim

to do the same comparison for OPLS3e and OPLS3. The data set includes the 8 systems originally reported in reference 1.

We also collected additional test series designed to

more stringently probe challenging chemical transformations akin to the outlier cases presented in Section 2.

The class of

perturbations most commonly associated with both the charge and torsion

transferability

issues

discussed

composition of conjugated ring systems.

involved

changes

in

This addition includes

the Chk1, Bace and FXa series first presented in Harder et al. Additional ligand series were sourced for FXa from references 3437,, Bace from reference 38 and another series focused on changes to an amidinium core, coordinating the catalytic aspartates of Bace

in

reference

19,

excluding

compounds

with

uncertain

protonation state or multiple possible tautomeric states. representative

ligand

for

each target is

shown

A

in Figure 4.

Additional data sets, representing a similar chemistry focus, was sourced from Schrodinger collaboration projects, denoted project A, B, C and D in Table 5.

In total 194 additional ligands were

added to the original 199 ligand benchmark. data

set

includes

various

targets

and

The final combined

compounds,

including

examples of kinases, proteases, tyrosine phosphatase, peptidase,


20

Page 21 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


positive/negative/neutral ligands, and variations of both cores and R-groups. The FEP+ protocol was similar to that previously described in reference 1, with a few notable differences. All simulations were run for 25ns, instead of 5ns, to improve convergence and enable us to

focus

on

force

field

related

incomplete loop was modeled in.

differences.

For

P38,

an

For FXa, after close inspection

of crystal structure densities, we noticed an ion binding site close to the ligand which was previously assigned as water.34,

36, 37

Consistent with the reported assay conditions, we replaced this water with a sodium ion.35,

37

We also refined ligand starting

geometries using a similar strategy to that discussed in reference 39.

When there was ambiguity about the pose, such as for symmetric

R-groups with meta or ortho substituents that could flip 180 degrees, we used FEP+ to assess and select the pose with the lowest free energy.

Starting pose coordinates for ligands and proteins

can be found in supplementary material. Associated PDB codes are provided in Table S2. The overall RMS error is improved by approximately 20% with OPLS3e.

That

improvement

is

primarily

improvement in the extended test series.

driven

by

accuracy

On the original FEP+

benchmark (rows 1-8), both OPLS3 and OPLS3e perform similarly well. The series that sees the greatest improvement using OPLS3e is ProjectC (OPLS3 RMSE = 2.2 kcal/mol, OPLS3e RMSE = 0.9 kcal/mol).


21


Page 22 of 53

The improvement is a product of both the improved torsions and charges.

An example perturbation improved by the OPLS3e torsions

is shown in Figure 5.

It involves the breaking of a fused ring

system in the core of the molecule leaving a dangling amide substituent.

Experimentally, breaking the fused ring results in

a 1.5 kcal/mol loss of potency.

In contrast, OPLS3 predicts a

loss of potency of 4.5 kcal/mol.

The simulated distribution of

the torsion connecting the amide substituent to the ring shown in the figure indicates the source of the error.

In both OPLS3 and

OPLS3e, the predominant state in the complex corresponds to a NCcc angle of approximately 180 degrees which situates the amide group in

the same plane as

the parent heterocycle

and

facilitates

formation of inter-molecule hydrogen bonds with the backbone of the protein.

The same conformation is the primary state sampled

by OPLS3e in solution.

However, in OPLS3, the primary state in

solution has the plane of the amide group orthogonal to the core heterocycle (NCcc=90).

Thus, the OPLS3 representation incorrectly

suggests that this variant will incur an additional conformational strain upon binding which leads to the observed under-prediction of its affinity versus the experimental reference data. Consistent with this trend, the QM and OPLS3e energy required to rotate this group to the NCcc=90 state is approx. 14 kcal/mol. OPLS3 under-estimates this energetic penalty by 7 kcal/mol which leads to the erroneous solvent population observed here.


Fixing

22

Page 23 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


just the torsion issues addresses a significant portion of the OPLS3 error over this series.

Using a model that combines OPLS3

charges with OPLS3e torsions reduces the series RMSE from 2.2 kcal/mol to 1.5 kcal/mol (Table 6).

The remaining improvement in

accuracy is due to the improved partial charge treatment with OPLS3e for the varied heterocyclic systems in this series. The set of amidine inhibitors of Bace-1 (Bace_cores) is another data set that shows significant improvement with OPLS3e.

The data

set includes the outlier case discussed in Section 2 and is comprised of a set of relatively uncommon heterocyclic cores (Figure 6) that form key interactions with catalytic aspartates in the binding site.

To accurately reflect how these variations

modulate the strength of their interaction with the catalytic aspartates, is a challenge for the force field.

The predictions

for this data set are relatively insensitive to the details of the torsion model and are primarily driven by changes in OPLS3e to the partial charges. The data set consists of two maps: one covering a series of monocyclic compounds (cmpd 1-10) and another covering a set of fused bicyclic compounds (cmpd 11-24).

For the monocyclic sub-

set the RMS error improves from 1.7 kcal/mol with OPLS3 to 1.3 kcal/mol.

The basis for the improvement is subtle and manifested

over the full data set, illustrating the benefit of OPLS3e for novel heterocycle design.

For the fused bicyclic subset the RMS


23


error improves from 2.2 kcal/mol to 1.5 kcal/mol.

Page 24 of 53

The improvement

is driven by better partial charge modeling of the aza derivatives in the series. unsubstituted

Table 7, gives the binding affinity for each

pyridyl

and

pyrmidyl

derivative

relative to the amino-isoindole core (cmpd 11).

in

the

series

As discussed in

Section 2, the poor OPLS3 predictions are a product of poor parameter transferability from the OPLS3 training set for the naryl substituted compounds. OPLS3e addresses this issue by fitting directly to QM leading to significantly better performance over this set. Despite these improvements the RMS error for the Bace_core data set (1.39 kcal/mol) remains quite high.

The remaining outlier,

which is mainly responsible for the residual error, is cmpd 23. As shown in Figure 7, relative to the amino-isoindole core (cmpd 11) cmpd 23 introduces an additional OMe group to the core ring. As indicated in Figure 7, the addition of the OMe substituent introduces a significant electron rich moiety in close proximity to ASP228 which leads to a significant drop in potency for cmpd 23.

However, both OPLS3 and OPLS3e significantly over-estimate

the associated drop in potency by approx. 3 kcal/mol.

In our FEP+

simulations the two catalytic aspartates in Bace-1 are assumed to be unprotonated in the holo state, in agreement with previous publications.19,

40

However, given the present observation we

conjectured that the introduction of substituents with significant


24

Page 25 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


electronegativity at this position may shift the preferred state of ASP228 to be neutral.

To test this hypothesis, we estimated

the pKa of this residue in the presence of both compounds #11 and #23 to be 4 and 7, respectively. SI Section 3.

The pKa protocol is discussed in

Given the associated experimental pH of the assay

is 4.5, this implies that the neutral form of ASP288 should be predominant for cmpd 23. Correcting the predicted binding affinity of cmpd #23 relative to cmpd #11, using Equation S6, gives +2.0 kcal/mol which is in-line with the experimentally measured value of +2.6 kcal/mol. We are presently pursuing strategies to include pKa effects, like this, in FEP+ calculations. Likewise, for the monocyclic molecules the binding affinity of core number 5 is still overestimated with OPLS3e. All Bace-1 cores in the data set are assumed to have a protonated amidine motif both in the protein binding site and in the reference state. Experimental core contribution to pKa is lower for core 5 than the other monocycles.19 The same is true for the calculated pKa, which for core 5 with two phenyl R-groups is 4.8,

close to the assay

conditions of pH=4.5. There is a possibility that core 5 is unprotonated before binding, which would result in an energetic penalty upon binding, and could explain the too high predicted affinity. In summary, there is still room for improvements, but overall OPLS3e

performs

well

for

all

tested

systems,


and

especially

25


Page 26 of 53

contributes a significant improvement to OPLS3 for challenging heterocycles. V.

Conclusion

We here report the development and validation of a new force field, OPLS3e, that leads to improved performance in predicting protein-ligand binding affinities.

Over an expansive set of

protein targets and ligand chemistries, the model achieves a RMS error of approx. 1 kcal/mol.

The model accomplishes this by

addressing limitations in chemotype transferability that affects small molecule torsions and charges.

Better torsion parameter

transferability is achieved with an expanded basis of torsion types that

distinguish

moieties

with

greater

chemical

specificity.

Better charge parameter transferability is achieved by directly fitting

ligand

specific

electrostatic potential.

charges

to

the

target

QM

derived

Ligand transformations that modulate the

composition of conjugated heterocycles are particularly sensitive to

these

effects

and

display

the

most

pronounced

accuracy

improvements. With regards to future development, we anticipate that additional parameter refinements to the present development (eg. adding new torsion types) will be necessary as more reference data becomes available.

We also expect that, as more aggressive explorations

of chemical and configurational space are enabled by continued advances in sampling efficiency, that will, in turn, expose new


26

Page 27 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


challenges to the fidelity of the force field.

In addition, we

are also actively looking into accuracy improvements that could be derived from modifications to the functional form of the model (eg. multi-dimensional dihedral terms, better VdW terms etc.) and improved treatment of regimes of chemical space that lie outside the

present

testing

data

(eg.

metalloenzymes).

Beyond

improvements to the force field, we are also actively pursuing strategies to account for the impact that changes in protonation state can have in FEP+ calculations. these

prospective

activities,

we

As a basis for prioritizing continue

to

investigate

outstanding prediction outliers to gain insights into the most important improvements that need to be made. Figures and Tables

Figure 1. relative

Left panel shows the experimental and FEP-predicted binding

free

energy

in

kcal/mol

distinguished by the ester-/amide-pyrrole groups.


between

ligands

The right panel 27


Page 28 of 53

shows the associated potential energy of the NCcn dihedral angle for an amide derivative derived from the FEP ligand (top) and the molecule used to train the associated torsion parameters (bottom). The

predominant

angle=180.

state

in

the

complex

corresponds

to

a

NCcn

The predominant state in solution corresponds to a

NCcn angle=0.

OPLS3* is a modified version of OPLS3 that re-

parameterizes the amide-pyrrole dihedral against

the fragment

analog of the amide-pyrrole substructure present in the left ligand.

Figure 2. Bace-1 inhibitor charge outliers. compound 11 in the binding site.

Left panel shows full

Bicyclic core variations with

OPLS3/QM amidine charges overlaid shown at right. Table summarizes the relative experimental and predicted binding free energy of the corresponding full compounds to Bace-1 in kcal/mol. The amidine


28

Page 29 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


motif forms key hydrogen bonds to the two catalytic aspartates of Bace-1. The predicted binding free energy for replacing one or two aromatic carbons for nitrogens (13, 16) in the amino-isoindole core (11) is consistently underestimated with OPLS3. Ligand series are

from

reference

19,

FEP+

calculations

employed

the

4DJW

structure. Table 1. RMS Error of with respect to QM for dimethyl aniline torsion profiles with varying para substituents (kcal/mol)a Para group

OPLS3 paramet er typea

OPLS3e paramet er typea

OPLS3 RMSE

OPLS3e RMSE

-H

A

A

0.0

0.0

-CH3

A

A

0.3

0.3

-F

A

B

0.6

0.0

-Cl

A

A

0.1

0.1

-NH2

A

C

1.4

0.0

-OH

A

D

1.2

0.1

-O-

A

E

4.1

0.2

-C#N

A

F

1.8

0.1

-Ph

A

A

0.3

0.2

a Letter symbols represent associated dimethyl aniline parameter type. All molecules share the same type in OPLS3 and the parameter is trained against -H. OPLS3e delineates the assignment for para -F, -NH2, -OH, -O- and -C#N functional groups.


29


Page 30 of 53

Figure 3. Rotamer energy comparisons for molecule pairs along the indicated dihedral.

Each pair (A-D) is a representative example

of the 4 chemical classes impacted by the OPLS3e torsion parameter type refinements.

In OPLS3, each pair shared the same parameter

assignment which now differ in OPLS3e. Table 2. Torsion profile RMS Error with respect to M06-2X/ccpVTZ(-f) (kcal/mol)a Model

all molecul

trainin g molecul

test molecule s: well


30

Page 31 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


es (25890)

es (20646)

covered (5244)

OPLS_200 5

1.9

1.9

1.8

OPLS3

1.2

1.2

1.3

OPLS3e

0.4

0.3

0.8

aNumber

of molecules for each set is shown in parentheses.

Table 3. Error summary for hydration free energy results, and transfer free energies to chloroform and toluene. Data set

No. cmpd s

RMSE (kcal/mol) OPLS3

OPLS3 + OPLS3e QM charges

418

1.02

1.39

0.98

Water-chloroform free energies

transfer 242

1.12

1.10

0.99

Water-toluene free energies

transfer 163

0.93

0.74

0.71

Hydration free energies

Table 4. Error summary for hydration free energies of chemical groups assigned high restraints in the new ligand based charge model. RMSE (kcal/mol) Chemical Group

No. cmpds

OPLS3

OPLS3 + OPLS3 QM e charges

Amines

55

1.39

2.17

1.30

Formates/este

49

0.89

2.04

1.16


31


Page 32 of 53

rs Nitros

29

1.05

1.36

0.80

Sulfoxides/su lfones/sulfat es/aryl thiols

5

0.66

2.45

0.76


32

Page 33 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4. Examples from series of compounds from the extended data set focused on variations of heterocycles. Table 5. Relative binding free energy results.


33


No.

No.

Page 34 of 53

RMSE (kcal/mol)

System

cmpd s

Seri es ref

BACE

36

41

58

1.11

1.10

CDK2

16

42

25

1.07

1.00

JNK1

21

43

34

0.81

0.88

MCL1

42

44

71

1.25

1.17

P38

34

45

56

0.89

0.88

PTP1B

23

46

49

0.70

0.63

Thrombin

11

47

16

0.85

1.08

Tyk2

16

48, 49

24

0.73

0.65

FXA

40

1417

60

1.47

1.29

CHK1

19

5055

18

1.20

0.98

BACE core 23

19

30

1.97

1.39

BACE heterocyc 21 les

38, 41

28

1.47

1.04

ProjectA

13

a

11

0.91

0.80

ProjectB

11

a

18

2.11

1.31

ProjectC

22

a

28

2.22

0.90

ProjectD

45

a

63

1.23

1.01

Total Weighted Average

393

589

1.29

1.04

Pert OPLS3 .

OPLS3 e

a In-house data.


34

Page 35 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 5. Left panel shows the experimental and FEP-predicted relative

binding

free

energy

in

kcal/mol

between

ligands

distinguished by the breaking of the fused lactam ring from the projectC series.

The right panel shows the simulated population

of the NCcc dihedral angle of the acyclic derivative. Table 6. ProjectC Relative Binding Free Energy Error Model

RMSE (kcal/m ol)

OPLS3

2.2

OPLS3 + 1.5 New Torsions OPLS3e

0.9


35


Page 36 of 53

Figure 6. Illustration of the chemical diversity of the challenging monocyclic and bicyclic core variations of Bace-1 inhibitors from reference 19. The description of the heterocycle variations and their effect on the key interactions were limited with OPLS3 and improved with OPLS3e. Table 7. Experimental and FEP Predicted Relative Binding Free Energies for Bace_core ligandsa Ligand perturbat iona

ddG (exp.)

ddG(OPLS3 )

ddG(OPLS3 e)


36

Page 37 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


11  12

0.7

5.3

2.3

11  13

0.0

4.0

1.0

11  14

0.2

4.0

0.5

11 15

0.5

4.5

1.6

11  16

0.9

3.3

0.9

a Specified ligands are illustrated in Figure 6. in kcal/mol.

Free energies

Figure 7. The experimental and FEP-predicted relative binding free energy in kcal/mol between ligands distinguished by the indicated substituent change (methoxy for hydrogen).

Substitution position

is in close proximity to ASP228. Tabulated FEP predictions are based on the receptor state indicated here where both ASP32 and ASP228 are modeled in their deprotonated form. ASSOCIATED CONTENT


37


Page 38 of 53

FEP results, input structures and OPLS3e parameters are included with the supplementary information. AUTHOR INFORMATION Corresponding Author * [email protected] Notes K.R. and C.W. contributed equally to this work.

The authors

declare the following competing financial interest(s): R.A.F. has a significant financial stake in, is a consultant for, and is on the Scientific Advisory Board of Schrodinger, Inc. K.R. was funded by the Swedish Research Council. REFERENCES (1)

Wang, L.; Wu, Y.; Deng, Y.; Kim, B.; Pierce, L.; Krilov,

G.; Lupyan, D.; Robinson, S.; Dahlgren, M. K.; Greenwood, J.; Romero, D. L.; Masse, C.; Knight, J. L.; Steinbrecher, T.; Beuming, T.; Damm, W.; Harder, E.; Sherman, W.; Brewer, M.; Wester, R.; Murcko, M.; Frye, L.; Farid, R.; Lin, T.; Mobley, D. L.; Jorgensen, W. L.; Berne, B. J.; Friesner, R. A.; Abel, R.

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.


38

Page 39 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(2)

Zwanzig, R. W. High‐Temperature Equation of State by a

Perturbation Method. I. Nonpolar Gases. J. Chem. Phys. 1954, 22, 1420–1426. (3)

Kollman, P. Free Energy Calculations: Applications to

Chemical and Biochemical Phenomena. Chem. Rev. 1993, 93, 2395–2417. (4)

Perez, A.; Morrone, J. A.; Simmerling, C.; Dill, K. A.

Advances in Free-Energy-Based Simulations of Protein Folding and Ligand Binding. Curr. Opin. Struct. Biol. 2016, 36, 25– 31. (5)

Hansen, N.; van Gunsteren, W. F. Practical Aspects of

Free-Energy Calculations: A Review. J. Chem. Theory Comput. 2014, 10, 2632–2647. (6)

Abel,

R.;

Wang,

L.;

Harder,

E.

D.;

Berne,

B.

J.;

Friesner, R. A. Advancing Drug Discovery through Enhanced Free Energy Calculations. Acc. Chem. Res. 2017, 50, 1625– 1632. (7)

Abel,

R.;

Mondal,

S.;

Masse,

C.;

Greenwood,

J.;

Harriman, G.; Ashwell, M. A.; Bhat, S.; Wester, R.; Frye, L.; Kapeller, R.; Friesner, R. A. Accelerating Drug Discovery through Tight Integration of Expert Molecular Design and


39


Page 40 of 53

Predictive Scoring. Curr. Opin. Struct. Biol. 2017, 43, 38– 44. (8)

Ciordia, M.; Pérez-Benito, L.; Delgado, F.; Trabanco, A.

A.; Tresadern, G. Application of Free Energy Perturbation for the Design of BACE1 Inhibitors. J. Chem. Inf. Model. 2016, 56, 1856–1871. (9)

Lovering, F.; Aevazelis, C.; Chang, J.; Dehnhardt, C.;

Fitz, L.; Han, S.; Janz, K.; Lee, J.; Kaila, N.; McDonald, J.; Moore, W.; Moretto, A.; Papaioannou, N.; Richard, D.; Ryan, M. S.; Wan, Z-K.; Thorarensen, A. Imidazotriazines: Spleen Tyrosine Kinase (Syk) Inhibitors Identified by FreeEnergy Perturbation (FEP). ChemMedChem 2015, 11, 217–233. (10)

Jorgensen, W. L. Computer-Aided Discovery of Anti-HIV

Agents. Bioorg. Med. Chem. 2016, 24, 4768–4778. (11)

Lenselink,

E.

B.;

Louvel,

J.;

Forti,

A.

F.;

van

Veldhoven, J. P. D.; de Vries, H.; Mulder-Krieger, T.; McRobb, F. M.; Negri, A.; Goose, J.; Abel, R.; van Vlijmen, H. W. T.; Wang, L.; Harder, E.; Sherman, W.; IJzerman, A. P.; Beuming, T. Predicting Binding Affinities for GPCR Ligands Using FreeEnergy Perturbation. ACS Omega 2016, 1, 293–304. (12)

Hünenberger

classical

P.

interaction

H.;

van

Gunsteren

functions

for


W.

F.

molecular

Empirical simulation

40

Page 41 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


BT

-

Computer

Theoretical

and

Simulation Experimental

of

Biomolecular

Applications.

In

Systems: Computer

Simulation of Biomolecular Systems; van Gunsteren W. F., Weiner P. K., Wilkinson A. J., Eds.; Springer: Dordrecht, Netherlands, 1997; 3, pp 3-82. (13)

Darden, T. A. Treatment of Long-Range Forces and Potential.

In Computational Biochemistry and Biophysics; Becker, O., MacKerell Jr. A.D., Roux, B., Watanabe, M., Eds.; CRC Press: Boca Raton, 2001. (14)

Halgren, T. A. Merck Molecular Force Field. I. Basis,

Form, Scope, Parameterization, and Performance of MMFF94. J. Comput. Chem. 2018, 17, 490–519. (15)

Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.;

Case, D. A. Development and Testing of a General Amber Force Field. J. Comput. Chem. 2004, 25, 1157–1174. (16)

Vanommeslaeghe, K.; Hatcher, E.; Acharya, C.; Kundu, S.;

Zhong, S.; Shim, J.; Darian, E.; Guvench, O.; Lopes, P.; Vorobyov, I.; Mackerell Jr. A. D. CHARMM General Force Field (CGenFF): A Force Field for Drug-like Molecules Compatible with the CHARMM All-Atom Additive Biological Force Fields. J. Comput. Chem. 2010, 31, 671–690. (17)

Mobley, D.; Bannan, C. C.; Rizzi, A.; Bayly, C. I.;

Chodera, J. D.; Lim, V. T.; Lim, N. M.; Beauchamp, K. A.;


41


Page 42 of 53

Slochower, D. R.; Shirts, M. R.; Gilson, M. K.; Eastman, P. K. Escaping Atom Types Using Direct Chemical Perception. J. Chem. Theory Comput. 2018, 14, 6076–6092. (18)

Harder, E.; Damm, W.; Maple, J.; Wu, C.; Reboul, M.;

Xiang, J. Y.; Wang, L.; Lupyan, D.; Dahlgren, M. K.; Knight, J. L.; Kaus, J. W.; Cerutti, D.; Krilov, G.; Jorgensen, W. L.; Abel, R.; Friesner, R. A. OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J. Chem. Theory Comput. 2016, 12, 281–296. (19)

Roos, K.; Viklund, J.; Meuller, J.; Kaspersson, K.;

Svensson,

M.

Potency

Prediction

of ß-Secretase

(BACE-1)

Inhibitors Using Density Functional Methods. J. Chem. Inf. Model. 2014, 54, 818–825. (20)

Roos, K.: Friesner, R. A.

Comparing Physics Based

Methods for Protein-Ligand Binding. Manuscript. (21)

Storer, J. W.; Giesen, D. J.; Cramer, C. J.; Truhlar, D.

G. Class IV Charge Models: A New Semiempirical Approach in Quantum Chemistry. J. Comput. Aided. Mol. Des. 1995, 9, 87– 110. (22)

Shivakumar, D.; Harder, E.; Damm, W.; Friesner, R. A.;

Sherman, W. Improving the Prediction of Absolute Solvation


42

Page 43 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Free Energies Using the Next Generation OPLS Force Field. J. Chem. Theory Comput. 2012, 8, 2553–2558. (23)

Tirado-Rives,

J.;

Jorgensen,

W.

L.

Contribution

of

Conformer Focusing to the Uncertainty in Predicting Free Energies for Protein−Ligand Binding. J. Med. Chem. 2006, 49, 5880–5884. (24)

Fateley, W. G.; Carlson, G. L.; Bentley, F. F. Phenolic-

OH Torsional Frequency as a Probe for Studying -Electron Distortions in Aromatic Systems. J. Phys. Chem. 1975, 79, 199–204. (25)

Pitt, W. R.; Parry, D. M.; Perry, B. G.; Groom, C. R.

Heteroaromatic Rings of the Future. J. Med. Chem. 2009, 52, 2952–2963. (26)

Rizzo, R. C.; Jorgensen, W. L. OPLS All-Atom Model for

Amines:

Resolution of the Amine Hydration Problem. J. Am.

Chem. Soc. 1999, 121, 4827–4836. (27)

Mobley, D. L. Experimental and Calculated Small Molecule

Hydration

Free

Pharmaceutical

Energies,

2013.

UC

Irvine:

Sciences,

Department

of

UCI.

http://www.escholarship.org/uc/item/6sd403pz (28)

Stephens, T. W.; Loera, M.; Quay, A. N.; Chou, V.; Shen,

C.; Wilson, A.; Acree Jr, W. E.; Acree, W. E.; Abraham, M. H.


43


Page 44 of 53

Correlation of Solute Transfer Into Toluene and Ethylbenzene from Water and from the Gas Phase Based on the Abraham Model. Open Thermodyn. J 2011, 5, 104–121. (29)

Sprunger, L. M.; Achi, S. S.; Acree, W. E.; Abraham, M.

H.; Leo, A. J.; Hoekman, D. Correlation and Prediction of Solute Transfer to Chloroalkanes from Both Water and the Gas Phase. Fluid Phase Equilib. 2009, 281, 144–162. (30)

Abraham, M. H.; Platts, J. A.; Hersey, A.; Leo, A. J.;

Taft, R. W. Correlation and Estimation of Gas − Chloroform and Water − Chloroform Partition Coefficients by a Linear Free Energy Relationship Method. J. Pharm. Sci. 1999, 88, 670–679. (31)

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. (32)

Shi, Y.; Xia, Z.; Zhang, J.; Best, R.; Wu, C.; Ponder,

J. W.; Ren, P. Polarizable Atomic Multipole-Based AMOEBA Force Field for Proteins. J. Chem. Theory Comput. 2013, 9, 4046–4063. (33)

Lopes, P. E. M.; Huang, J.; Shim, J.; Luo, Y.; Li, H.;

Roux, B.; MacKerell, Jr. A. D. Force Fields for Peptides and


44

Page 45 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Proteins based on the Classical Drude Oscillator. J. Chem. Theory Comput. 2013, 9, 5430–5449. (34)

Yoshikawa, K.; Kobayashi, S.; Nakamoto, Y.; Haginoya,

N.; Komoriya, S.; Yoshino, T.; Nagata, T.; Mochizuki, A.; Watanabe, K.; Suzuki, M.; . Design, Synthesis, and SAR of Cis-1,2-Diaminocyclohexane Derivatives as Potent Factor Xa Inhibitors.

Part

II:

Exploration

of

6–6

Fused

Rings

as

Alternative S1 Moieties. Bioorg. Med. Chem. 2009, 17, 8221– 8233. (35)

Yoshikawa, K.; Yokomizo, A.; Naito, H.; Haginoya, N.;

Kobayashi, Osanai,

S.;

K.;

Yoshino,

Watanabe,

T.;

K.;

Nagata,

Kanno,

T.;

H.;

Mochizuki,

Ohta,

T.

A.;

Design,

Synthesis, and SAR of Cis-1,2-Diaminocyclohexane Derivatives as Potent Factor Xa Inhibitors. Part I: Exploration of 5-6 Fused Rings as Alternative S1 Moieties. Bioorganic Med. Chem. 2009, 17, 8206–8220. (36)

Zbinden, K. G.; Anselm, L.; Banner, D. W.; Benz, J.;

Blasco, F.; Décoret, G.; Himber, J.; Kuhn, B.; Panday, N.; Ricklin, F.; Risch, P.; Schlatter, D.; Stahl, M.; Thomi, S.; Unger, R.; Haap, W. Design of Novel Aminopyrrolidine Factor Xa Inhibitors from a Screening Hit. Eur. J. Med. Chem. 2009, 44, 2787–2795.


45


(37)

Page 46 of 53

Chan, C.; Borthwick, A. D.; Brown, D.; Burns-Kurtis, C.

L.; Campbell, M.; Chaudry, L.; Chung, C-w.; Convery, M. A.; Hamblin, J. N.; Johnstone, L.; Kelly, H. A.; Kleanthous, S.; Patikis, A.; Patel, C.; Pateman, A. J.; Senger, S.; Shah, G. P.; Toomey, J. R.; Watson, N. S.; Weston, H. E.; Whitworth, C.; Young, R. J.;

Zhou, P. J. Factor Xa Inhibitors: S1

Binding Interactions of a Series of N-{(3S)-1-[(1S)-1-Methyl2-Morpholin-4-Yl-2-Oxoethyl]-2-Oxopyrrolidin-3Yl}Sulfonamides. J. Med. Chem. 2007, 50, 1546–1557. (38)

Truong, A. P.; Probst, G. D.; Aquino, J.; Fang, L.;

Brogley, L.; Sealy, J. M.; Hom, R. K.; Tucker, J. A.; John, V.; Tung, J. S.; Pleiss, M. A.; Konradi, A. W.; Sham, H. L.; Dappen, M. S.; Toth, G.; Yao, N.; Brecht, E.; Pan, H.; Artis, D. R.; Lany, R.; Bova, M. P.; Sinha, S.; Yednock, T. A.; Zmolek,

W.;

Permeability

Quinn,

K.

of

Hydroxyethylamine

the

P.;

Sauer,

J-M.

Improving

BACE-1

the

Inhibitors:

Structure–Activity Relationship of P2′ Substituents. Bioorg. Med. Chem. Lett. 2010, 20, 4789–4794. (39)

Kaus, J. W.; Harder, E.; Lin, T.; Abel, R.; McCammon, J.

A.; Wang, L. How To Deal with Multiple Binding Poses in Alchemical

Relative

Protein–Ligand

Binding

Free

Energy

Calculations. J. Chem. Theory Comput. 2015, 11, 2670–2679.


46

Page 47 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(40)

Swahn, B.; Holenz, J.; Kihlström, J.; Kolmodin, K.;

Lindström, J.; Plobeck, N.; Rotticci, D.; Sehgelmeble, F.; Sundström, M.; Berg, S. Von; Falting, J.; Georgievska, B.; Gustavsson, S.; Neelissen, J.; Ek, M.; Olsson, L-L.; Berg, S. Aminoimidazoles

as

BACE-1

Inhibitors :

The

Challenge

to

Achieve in Vivo Brain Efficacy. Bioorg. Med. Chem. Lett. 2012, 22, 1854–1859. (41)

Cumming, J. N.; Smith, E. M.; Wang, L.; Misiaszek, J.;

Durkin, J.; Pan, J.; Iserloh, U.; Wu, Y.; Zhu, Z.; Strickland, C.; Voigt, J.; Chen, X.; Kennedy, M. E.; Kuvelkar, R.; Hyde, L. A.; Cox, K.; Favreau, L.; Czarniecki, M. F.; Greenlee, W. J.;

McKittrick,

B.

A.;

Parker,

E.

M.;

Stamford,

A.

W.

Structure Based Design of Iminohydantoin BACE1 Inhibitors: Identification of an Orally Available, Centrally Active BACE1 Inhibitor. Bioorg. Med. Chem. Lett. 2012, 22, 2444–2449. (42)

Hardcastle, I. R.; Arris, C. E.; Bentley, J.; Boyle, F.

T.; Chen, Y.; Curtin, N. J.; Endicott, J. A.; Gibson, A. E.; Golding, B. T.; Griffin, R. J.; Jewsbury, P.; Menyerol, J.; Mesguiche, V.; Newell, D. R.; Noble, M. E.; Pratt, D. J.; Wang,

L.

Z.;

Whitfield,

H.

J.

Cyclohexylmethylguanine Derivatives:

N2-Substituted

O6-

Potent Inhibitors of

Cyclin-Dependent Kinases 1 and 2. J. Med. Chem. 2004, 47, 3710–3722.


47


(43)

Page 48 of 53

Szczepankiewicz, B. G.; Kosogof, C.; Nelson, L. T. J.;

Liu, G.; Liu, B.; Zhao, H.; Serby, M. D.; Xin, Z.; Liu, M.; Gum, R. J.; Haasch, D. L.; Wang, S.; Clampit, J. E.; Johnson, E. F.; Lubben, T. H.; Stashko, M. A.; Olejniczak, E. T.; Sun, C.; Dorwin, S. A.; Haskins, K.; Abad-Zapatero, C.; Fry, E. H.;

Hutchins,

C.

W.;

Sham,

H.

L.;

Rondinone,

C.

M.;

Trevillyan, J. M. Aminopyridine-Based c-Jun N-Terminal Kinase Inhibitors with Cellular Activity and Minimal Cross-Kinase Activity. J. Med. Chem. 2006, 49, 3563–3580. (44)

Friberg, A.; Vigil, D.; Zhao, B.; Daniels, R. N.; Burke,

J. P.; Garcia-Barrantes, P. M.; Camper, D.; Chauder, B. A.; Lee, T.; Olejniczak, E. T.; Fesik, S. W. Discovery of Potent Myeloid Cell Leukemia 1 (Mcl-1) Inhibitors Using FragmentBased Methods and Structure-Based Design. J. Med. Chem. 2013, 56, 15–30. (45)

Goldstein, D. M.; Soth, M.; Gabriel, T.; Dewdney, N.;

Kuglstatter, A.; Arzeno, H.; Chen, J.; Bingenheimer, W.; Dalrymple, S. A.; Dunn, J.; Farrell, R.; Frauchiger, S.; La Fargue, J.; Ghate, M.; Graves, B.; Hill, R. J.; Li, F.; Litman, R.; Loe, B.; McIntosh, J.; McWeeney, D.; Papp, E.; Park, J.; Reese, H. F.; Roberts, R. T.; Rotstein, D.; San Pablo, B.; Sarma, K.; Stahl, M.; Sung, M.-L.; Suttman, R. T.; Sjogren, E. B.; Tan, Y.; Trejo, A.; Welch, M.; Weller, P.;


48

Page 49 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Wong, B. R.; Zecic, H. Discovery of 6-(2,4-Difluorophenoxy)2-[3-Hydroxy-1-(2-Hydroxyethyl)Propylamino]-8-Methyl-8HPyrido[2,3-d]Pyrimidin-7-One

(Pamapimod)

and

6-(2,4-

Difluorophenoxy)-8-Methyl-2-(Tetrahydro-2H-Pyran-4Ylamino)Pyrido[2,3-d]Pyrimidin-7(8H)-One (R1487) as Orally Bioavailible and Highly Selective Inhibitors of p38 MitogenActivated Protein Kinase. J. Med. Chem. 2011, 54, 2255–2265. (46)

Wilson, D. P.; Wan, Z.-K.; Xu, W.-X.; Kirincich, S. J.;

Follows, B. C.; Joseph-Mccarthy, D.; Foreman, K.; Moretto, A.; Wu, J.; Zhu, M.; Binnun, E.; Zhang, Y.-L.; Tam, M.; Erbe, D. V.; Tobin, J.; Xu, X.; Leung, L.; Shilling, A.; Tam, S. Y.; Mansour, T. S.; Lee, J. Structure-Based Optimization of Protein Tyrosine Phosphatase 1B Inhibitors:

From the Active

Site to the Second Phosphotyrosine Binding Site. J. Med. Chem. 2007, 50, 4681–4698. (47)

Baum, B.; Mohamed, M.; Zayed, M.; Gerlach, C.; Heine,

A.; Hangauer, D.; Klebe, G. More than a Simple Lipophilic Contact:

A

Detailed

Thermodynamic

Analysis

of

Nonbasic

Residues in the S1 Pocket of Thrombin. J. Mol. Biol. 2009, 390, 56–69. (48)

Liang, J.; Tsui, V.; Van Abbema, A.; Bao, L.; Barrett,

K.; Beresini, M.; Berezhkovskiy, L.; Blair, W. S.; Chang, C.; Driscoll,

J.;

Eigenbrot,

C.;

Ghilardi,


N.;

Gibbons,

P.; 49


Page 50 of 53

Halladay, J.; Johnson, A.; Kohli, P. B.; Lai, Y.; Liimatta, M.; Mantik, P.; Menghrajani, K.; Murray, J.; Sambrone, A.; Xiao, Y.; Shia, S.; Shin, Y.; Smith, J.; Sohn, S.; Stanley, M.; Ultsch, M.; Zhang, B.; Wu, L. C.; Magnuson, S. Lead Identification of Novel and Selective TYK2 Inhibitors. Eur. J. Med. Chem. 2013, 67, 175–187. (49)

Liang, J.; van Abbema, A.; Balazs, M.; Barrett, K.;

Berezhkovsky, L.; Blair, W.; Chang, C.; Delarosa, D.; DeVoss, J.; Driscoll, J.; Eigenbrot, C.; Ghilardi, N.; Gibbons, P.; Halladay, J.; Johnson, A.; Kohli, P. B.; Lai, Y.; Liu, Y.; Lyssikatos, J.; Mantik, P.; Menghrajani, K.; Murray, J.; Peng, I.; Sambrone, A.; Shia, S.; Shin, Y.; Smith, J.; Sohn, S.; Tsui, V.; Ultsch, M.; Wu, L. C.; Xiao, Y.; Yang, W.; Young,

J.;

Optimization

Zhang, of

a

B.;

Zhu,

B.-y.;

4-Aminopyridine

Magnuson,

Benzamide

S.

Lead

Scaffold

To

Identify Potent, Selective, and Orally Bioavailable TYK2 Inhibitors. J. Med. Chem. 2013, 56, 4521–4536. (50)

Lin, N.-H.; Xia, P.; Kovar, P.; Park, C.; Chen, Z.;

Zhang,

H.;

Rosenberg, S. H.;

Sham,

H.

L. Synthesis and

Biological Evaluation of 3-Ethylidene-1,3-Dihydro-Indol-2Ones as Novel Checkpoint 1 Inhibitors. Bioorg. Med. Chem. Lett. 2006, 16, 421–426.


50

Page 51 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(51)

Huang, S.; Garbaccio, R. M.; Fraley, M. E.; Steen, J.;

Kreatsoulas, C.; Hartman, G.; Stirdivant, S.; Drakas, B.; Rickert, K.; Walsh, E.; Hamilton, K.; Buser, C. A.; Hardwick, J. Mao, X.; Abrams, M.; Beck, S.; Tao, W.; Lobell, R.; Laura Sepp-Lorenzino,

L.;

Yan,

Y.;

Ikuta,

M.;

Murphy,

J.

Z.;

Sardana, V.; Munshi, S.; Kuo, L.; Reillyd, M.; Mahand, E. Development of 6-Substituted Indolylquinolinones as Potent Chek1 Kinase Inhibitors. Bioorg. Med. Chem. Lett. 2006, 16, 5907–5912. (52)

Oza, V.; Ashwell, S.; Almeida, L.; Brassil, P.; Breed,

J.; Deng, C.; Gero, T.; Grondine, M.; Horn, C.; Ioannidis, S.; Liu, D.; Lyne, P.; Newcombe, N.; Pass, M.; Read, J.; Ready, S.; Rowsell, S.; Su, M.; Toader, D.; Vasbinder, M.; Yu, D.; Yu, Y.; Xue, Y.; Zabludoff, S.; Janetka, J. Discovery of

Checkpoint

Kinase

Inhibitor

(S)-5-(3-Fluorophenyl)-N-

(Piperidin-3-Yl)-3-Ureidothiophene-2-Carboxamide by

Structure-Based

Design

and

(AZD7762)

Optimization

of

Thiophenecarboxamide Ureas. J. Med. Chem. 2012, 55, 5130– 5142. (53)

Zhao, L.; Zhang, Y.; Dai, C.; Guzi, T.; Wiswell, D.;

Seghezzi, W.; Parry, D.; Fischmann, T.; Siddiqui, M. A. Design, Synthesis and SAR of Thienopyridines as Potent CHK1 Inhibitors. Bioorg. Med. Chem. Lett. 2010, 20, 7216–7221.


51


(54)

Page 52 of 53

Tao, Z-F.; Wang, L.; Stewart, K. D.; Chen, Z.; Gu, W.;

Bui, M-H.; Merta, P.; Zhang, H.; Kovar, P.; Johnson, E.; Park, C.; Judge, R.; Rosenberg, S.; Sowin, T.; Lin, N-H. StructureBased Design, Synthesis, and Biological Evaluation of Potent and Selective Macrocyclic Checkpoint Kinase 1 Inhibitors. J. Med. Chem. 2007, 50, 1514–1527. (55)

Huang, X.; Cheng, C. C.; Fischmann, T. O.; Duca, J. S.;

Yang, X.; Richards, M.; Shipps, G. W. Discovery of a Novel Series of CHK1 Kinase Inhibitors with a Distinctive Hinge Binding Mode. ACS Med. Chem. Lett. 2012, 3, 123–128.


52

Page 53 of 53 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table of Contents Graphic OPLS3e:

Extending

Force

Field

Coverage

for

Drug-Like

Small

Molecules Katarina Roos, Chuanjie Wu, Wolfgang Damm, Mark Reboul, James M. Stevenson, Chao Lu, Markus K. Dahlgren, Sayan Mondal, Wei Chen, Lingle Wang, Robert Abel, Richard A. Friesner, Edward Harder


53

OPLS3e: Extending Force Field Coverage for Drug-Like Small

Recommend Documents