This scientific roadmap includes the next two planned force field releases and a list of scientific studies which need to be performed in 2020. Each study has a priority assigned to it. This roadmap will be continuously updated.

Force Fields

Upcoming force field versions:

Version

Codename

Features

Expected release date

Comment / Blocker

openff-1.2.0

Parsley

  • Redesigned QM dataset for parameterization with better/broader coverage

  • Parameter fixes

Expected: May 2020

Released: Jun 3, 2020

openff-1.3.0

Parsley

  • Improvements to priors by element, etc., allowing for more accurate fitting

  • Filtering of molecules used in fitting certain torsions (esp. for amides) to improve parameter quality (came up in the progress of fixing a problem elsewhere)

  • Discussed in this release call: 2020-09-17 FF Release Meeting notes

When ready. September-December 2020.

openff-2.0.0

Sage

  • LJ refit (based on the ongoing feasibility study)

  • Limited WBO torsion interpolation for systems for which data already exists (more torsional data needed for a wide range application)

Exepected: Release date to be set by November 2020 and will have three month lead time. Likely release between December 2020 and Feb. 2021.

  • How fast can we do WBO interpolations (Pavan)

  • Simon Boothroyd needs to get in touch with David Hahn and folks from the Chodera lab to discuss some free energy benchmarking after LJ fitting

  • Late 2020/early 2021 still feasible

Scientific studies

The list of scientific studies which need to be performed in 2020, which will be updated every 3 months, as suggested in the science project management workflow. Each study should be linked to its Confluence page with more information about study design, execution and results. The study design should be submitted before study is about to begin.

Estimate start dates and end dates when possible before study has started. Record the real start and end dates for each study below the estimated dates.

Labels

Category

Labels

Priority

| |

Effort

| |

Status

| | | | |

Study

Priority

Effort

Science dependencies

Infrastructure dependencies

Comment

Start date

End date

Status

Driver/Team

Chemical perception

Addition of new parameters – manually fixing problems

Made easier by benchmarking dashboard (Optional)

Made easier by benchmarking dashboard (Optional

Hyesu Jang David Mobley Jessica Maat (Deactivated) Victoria Lim (Deactivated)

Automated typing inference from scratch

Organise a meeting to coordinate efforts.

Update: Tobias Huefner is doing some basic research, but we don’t have a timeline defined here. Perhaps a more specific study to look at typing issues similar to Schauperl’s work on LJ typing.

Slowly

Full-time person needed – to be discussed further. Work of Josh Fass (Deactivated) and Tobias Huefner may assist here. Owen Madin interested. Trevor Gokey is also actively working in this area.

Mixture Properties

Binary Mixture Data Feasibility Study

In the writing stage.

Driver: Simon Boothroyd
Team: Michael Shirts Owen Madin

Non-bonded optimization

Parent study for in a long-term progress stage.

Driver:Simon Boothroyd
Team: Michael Shirts Owen Madin

Chemical potential-like properties

Non-bonded optimization

Implementation in Evaluator

Need to evaluate the data first (testing needed). Add Confluence page here.

Simon Boothroyd

Solvent-solvent partition coefficients

Implementation in Evaluator

  • Data needed, harder problem.

  • Update: Access to solubility phase, data is less of a problem now (MNSOL)

Simon Boothroyd

Data coverage and availability

Feasibility studies

Check the available data and identify missing data points. Worry in the future what to do about it. We will use what we have for Sage.

Ongoing

Simon Boothroyd Owen Madin Michael Shirts

QM Data Generation

QM dataset selection (training data) for OpenFF-1.2.0

Need to expand to benchmarking set.

David Mobley Jessica Maat (Deactivated) Hyesu Jang

QM dataset selection for OpenFF-2.0.0

David Mobley Jessica Maat (Deactivated) Hyesu Jang
Lead: Hyesu Jang ??

Benchmarking/re-evaluating our choice of QM theory

(Optional) QC Dataset submission infrastructure

Test of the whole torsiondrive. Keep within 10-50 torsiondrives. More is better.

  • Some datasets ready, but analysis is still required (Hyesu Jang ) Pavan might help with this.

  • No changes made to our fitting data til after 2.0 release, but decision may be made before that (eg while fitting)

Hyesu Jang lead; Lee-Ping Wang Pavan Behara

Hyesu Jang also leading molecule set selection with help from Jessica Maat (Deactivated) and Victoria Lim (Deactivated)

Protomer/tautomer enumerated molecules

QM level of theory validation (QMLoTV)

Protonation/tautomer enumeration integration (Joshua Horton doing OE version in toolkit; there’s currently no good protonation state enumeration with RDKit – see )

  • It can only do enumeration with OpenEye

Joshua Horton

Data selection for ionic species

What kind of experimental data would we need to include charged molecules?

Lead: Simon Boothroyd (oversight, to assemble team?)

Data on molecules with nonzero formal charges

QM level of theory validation (QMLoTV)

(Optional) QC Dataset submission infrastructure

Couples to benchmarking

Pavan Behara

Enamine REAL fragment coverage

Automated fragmentation integration Joshua Horton

Trevor Gokey

Ligand Expo fragment coverage

Automated fragmentation integration Joshua Horton

Ligand Expo has higher priority than Enamine Real.

Richer torsion data for WBO fitting

WBO torsion implementation

  • What data to generate and

(person needed to continue work of Chaya Stern (Deactivated); Will be Pavan Behara with Jessica Maat (Deactivated) , eventually coordinating with Simon Boothroyd as he moves to his new position. )

Biopolymer data selection (ensure sidechain data is available in QCA)

One dataset ready, but a lot more data needs to be generated if we want sidechain sampling

David Cerutti (Deactivated)

Biopolymer data computation

(Optional) QC Dataset submission infrastructure

David Cerutti (Deactivated) David Dotson

More efficient torsion sampling with less grid points during scan

Fitting

Addition of new parameters – manually fixing problems

Ongoing

Hyesu Jang David Mobley Jessica Maat (Deactivated) Victoria Lim (Deactivated)

LJ refitting (Sage)

Non-bonded optimization

Simon Boothroyd and Owen Madin

WBO refitting (Sage)

More torsion data

WBO torsion implementation: Done.

Implement what Chaya has already done. Infrastructure ready.

Pavan Behara and Jessica Maat (Deactivated) to divide up chemical space for fitting/test fixes. Hyesu Jang will do central fitting of production FF. Additional discussion: https://openforcefield.atlassian.net/wiki/spaces/OFFO/pages/670629936/90-day+plan+Onboarding+for+Pavan+Behara?accessType=view&grantAccess=true&username=5f592e4d0b6be2006ef56679&userFullName=Pavan%20Behara#Notes-from-9%2F16%2F20-planning-meeting-(feel-free-to-migrate)

After May meeting

Late 2020 (Sep 2020)

Jessica Maat (Deactivated) and Pavan Behara . Pavan Behara will eventually oversee.

BCC refitting

LJ refit

Patterns for BCCs; could start with something simple like bond SMARTS.

ChargeIncrementModel implementation (early May)

Simon Boothroyd Owen Madin

Study how to set prior widths and weights for different sorts of data during FF optimization

Became higher priority due to need for fixes. Headed towards a 1.3 release late 2020.

Lee-Ping Wang Hyesu Jang Spinoff?

Value of data generated “incidentally” during torsiondrive in fitting, e.g. optimization snapshots, gradients, energies (low control over these data points)

Some parts of Bespoke workflow

  • Once we have more people working on fitting, someone can run this study

Joshua Horton (question)

Benchmarking

Small reference system for fast testing of FE infrastructure – 5-10 small reference systems, possibly subset of SAMPL challenges, for comparison of different free energy methods to avoid using large P-L systems for test calculations

Should use SAMPLing challenge systems plus a couple more similar ones.

ASAP

Benchmarking/re-evaluating our choice of QM theory

See above; to be done while fitting 2.0 (datasets ready). Hyesu Jang leading.

Lee-Ping Wang Hyesu Jang Pavan Behara

CCDC data selection/release

Create a list of tests to judge the “quality” of biopolymer FF with our scientific advisory board

Organise the meeting with our IAB, invite to May meeting. Done.

DC and MS will start conversations to get this going.

April / May

David Cerutti (Deactivated) Michael Shirts

openff-1.2.0 (Parsley) benchmarking

Minor release of Parsley

Benchmarking dashboard

Done in preprint form, but no benchmarking dashboard. Still need torsion benchmarking; utilize work just done for OpenFF 1.0 paper.

JDC is trying to get a complete FE set run by D. Rufa.

Mid 2020

Done-ish

openff-2.0.0 (Sage) benchmarking

Release of Sage

Benchmarking dashboard

Late 2020

Biopolymers

Which quantum method should we use for biopolymers (should it be the same as small molecules)?

QM benchmarking study

Short term – using the same method and same level of theory as ANI (wB97D)

Lee-Ping Wang David Cerutti (Deactivated)

Feasibility/benchmarking studies of torsional CMAPs

After protein FF implementation

CMAP support in OFFTK

David Cerutti (Deactivated)

Feasibility/benchmarking studies of other cross-terms

Support for cross-terms in OFFTK

MS – Importance of cross-terms will be related to a number of types

Charges

GCN charge model

In a few steps:

  • conda-installable tool to assign charges

  • integration of tool into OFFTK under ChargeIncrementModel keyword (and exposure of relevant keywords)

John Chodera Yuanqing Wang Josh Fass (Deactivated) (maybe John Herr)

Off-site charge SMIRKS definition/fitting/benchmarking

VirtualSite support in OFFTK

Helpful discussion in Slack: https://openforcefieldgroup.slack.com/archives/C1907SGET/p1590251452068100

Infrastructure expected in September 2020

(but interface with David Cerutti (Deactivated) work?)

Bayesian inference and surrogate modeling

Testing Bayesian inference on an analytical model

Nearing completion

Owen Madin

Generalizing analytical model for Bayesian inference and testing methods

Proof-of-concept work to give us an analytical form for early testing

slower

Owen Madin (and a student)

Constructing full Bayesian architecture with reweighting and simulation to build surrogate models

Analytical Bayesian inference testing

ForceBalance → pytorch, torchMD (timemachine)

John Herr (question)

Owen Madin (science, not software)

Automated typing inference from scratch

Bayesian-based typing (Josh Fass’s work)

Josh Fass (Deactivated)Tobias Huefner

Other

Water co-optimization planning study (to be executed later) – discuss with Lee-Ping Wang

Lack of bandwidth, potentially Bill Swope could help advise with data selection.

Thinking about metals / ions / salts

Biologically relevant, will become high priority at some point

Thinking about ionic liquids

Alchemical force fields (for alchemical free energies)

/

Soft core potentials. JDC might have people in his lab working on it, MS is interested to join the effort.

Continuous (smearnoff) typing

ESPALOMA

Yuanqing Wang