What's up with OpenFF?
- James Eastwood
- Matt Thompson
- Jeffrey Wagner
The past few months have been somehow both busy and quiet for the staff of OpenFF. We’ve said goodbye to some old colleagues and hello to new. We’ve been improving our infrastructure, pioneering new science, and improving our internal work processes. We’ve made big advances in our interoperability with other molecular software packages. We’ve advanced an AI model for a crucial force field component (electrostatics). We have developed a new and improved workflow for parameterizing covalently-modified proteins. We’ve been actively supporting researchers from the broader OpenFF Initiative in some of their most valuable projects. And we gathered with many of our colleagues for an in-person work week in Irvine, CA! Here are some highlights from the past six months.
Scientific advances
Our flagship product here at Open Force Field is, of course, force fields, and we continue to make improvements to the Sage force field line. The latest release, v2.2.1, came out in September and fixed some angles that should be linear so that they remain closer to 180.
Our science team has been hard at work on an AI tool to assign partial charges to molecules. The software tool, OpenFF-NAGL, is trained to reproduce the AM1-BCC charges that our current force fields use. A release candidate NAGL model is now available for testing, and is nearly ready for use in fitting a new force field. Meanwhile, our colleague Meghan Osato, a student in Prof. David Mobley’s lab at UC Irvine, has done some great work exploring the limitations of conventional AM1-BCC partial charges, and the impact that they have on free energy calculations. Her work, available now as a pre-print, underscores the importance of neural network charges, which avoids these sources of error.
Many of our users are interested in simulating proteins that include covalent bonds to things that aren’t canonical amino acids. These systems might be cyclic peptides with synthetic side chains, protein-drug conjugates, or natural post-translational modifications. We have developed a prototype workflow for setting up and parameterizing these kinds of systems, and we’d love feedback on your use cases and the shortcomings of this prototype to help us develop this into a more useful feature of our public API moving forward.
Supporting force field developments
The structure of our project puts most of our core infrastructure and incremental science under the care of the OpenFF Consortium-funded staff at OMSF, while many of the scientific advances come from our academic collaborators. The staff have an important role to play in supporting these scientific advances.
Chapin Cavender, working in Prof. Mike Gilson’s lab at UCSD, has been hard at work developing a force field with self-consistent parameters for both proteins and small molecules. Our staff has been working closely with him, helping him set up his fitting workflow with more computing power and running some parallel experiments to help advance his research.
We have been helping students from Prof. Michael Shirts’ lab at CU Boulder running fitting experiments on proteins and lipids, and developing workflows for the parameterization of ion force fields in complex aqueous mixtures.
Infrastructure improvements
Our infrastructure team has recently made some big improvements to our interoperability backend, Interchange. Among many other improvements, we have improved support for importing and exporting files for use with GROMACS.
Many useful workflows for setting up systems with our force fields are limited by the ability to load a PDB file. After testing out a few different software tools for this purpose, we realized we would have to write our own. A early version of this PDB loading tool is available as openff-pablo, and can be seen in the PTM workflow example above.
We also have improved our internal benchmarking tool, yammbs, so that benchmarking a new candidate force field is close to a push-button process.
Our users have a very wide set of ideas for what to do with our force fields, so to facilitate the exchange of these crazy ideas, we launched a new repo, anything-goes. We make no guarantees about the contents of anything in there, and invite anyone to submit their scripts that show off unusual uses of OpenFF tools. Our attitude to contributions on this repo is that PRs are “default approved”, provided they don’t have malicious code or big files that would inflate the size of the git repo.
Staff Changes
In the fall, we said goodbye to two staff members whose grant-funded positions came to an end: Alexandra McIsaac and Brent Westbrook. They both moved on to exciting new positions that match their expertise well, and we wish them the best! Around the same time, we welcomed Jennifer Clark, who has taken responsibility for our quantum chemical dataset strategy. She has already made a big impact on our work, and we are excited to see what she accomplishes next!
Internal Processes
Anyone who has met Zachary Baker, the community manager at OMSF, probably knows he’s a true believer in the Agile approach to project management. He has been helping us improve our processes for tracking, prioritizing, and assigning work. One of the most exciting changes to come out of this effort has been getting the whole team into a daily 15-minute standup meeting! Since our team members are spread from New York to Canberra and many places in between, this has required some flexibility in our calendars, but the time with all on the same video call has been invaluable.
Meeting In Person
The structure of our organization, where staff work remotely from all over the world and we coordinate closely with many different academic research groups, means that it is hard for everyone in OpenFF to know what everyone else is doing and develop a shared vision of where we are, where we are going, and what steps to take next across our many projects. To address this need, we set aside an entire work week in February to gather in Irvine, CA. We were joined in Irvine by researchers from several different academic groups, and some of our collaborators from the UK joined us by video conference. Huge thanks to Prof. David Mobley for hosting us all and shuttling everyone around in his van!
