OMSF Newsletter - Summer 2025

Hello from the Open Molecular Software Foundation!

As summer sets in, we’re closing out another busy and inspiring quarter across the OMSF ecosystem. From milestone releases to community growth, new hires to new infrastructure, the past few months have been all about building momentum. Read on to catch up on the latest across our projects, platforms, and people.

We’ve welcomed a new face to our org -  Cynthia Xu joined the OpenADMET team back in June. With an incredible background in bioanalytics, she’s been tackling the hard science that powers OpenADMETs predictive modeling. She’s been great!

Applications for the next OMSF Biosecurity Fellow, a position previously held by Sam Curtis, have closed! We will announce Sam's successor shortly - thanks to the Rosetta team for funding this important role.

Our OMSF Blog is up and running (you might be here!), and has proven to be a fantastic way to stay connected with our staff and community members. With research topics, individual updates, and opinions on best practices, the OMSF blog is a must-follow for anyone looking to do open science in molecular modeling. Our board president, David Mobley, wrote a great piece a while back - read it as well as all the others here!

If you ever wanted a map of our repositories and software products, here it is! We’ve added an OMSF Software Directory to our website! Here you can find all the information about packages, tools, and workflows developed by OMSF projects. It also includes other open source packages - if you’d like to add your repo, the OMSF Directory makes it easy! Big shout out to Ethan Holz for the effort!

And if you missed it - the OMSF Playbooks are live! A set of living, practical handbooks detailing the tried and true methods used by our organization, these playbooks turn our best practices into step-by-step recipes. We currently have our Developer, Benchmarking, and Contributions playbooks for you to enjoy, with more on the way. From picking a repo layout to testing plans to catching benchmarking regressions early on, these are full of information. Check the link above to find out more!

Project Highlights

OpenFold

It’s been a transformative quarter for OpenFold. Development on OpenFold3 has accelerated with the completion of a new inference pipeline, enabling users to input a sequence and receive a structure with a ligand bound - an MVP system designed for end-to-end usability. This pipeline is undergoing final packaging with support from the Eco Infra team, our internal OMSF developers, and will be distributed in a containerized Docker image for easy installation and deployment. We sent out a beta version of OpenFold3 using an early checkpoint to members and employees of OpenFold3, and our public release is expected in the coming months.

The latest OpenFold3 training checkpoint shows promising results. Preliminary benchmarks suggest it outperforms other models currently available for protein/ligand predictions, although some fine-tuning remains. Work on the user interface is underway, and feedback from early testers within the consortium is helping us shape the experience. If you’d like access to the beta release, think about joining the OpenFold consortium!

To that end, OpenFold’s community continues to grow, with membership reaching more than 30 organizations. These members include both biotech and pharmaceutical companies like Johnson & Johnson, Apheris, Dyno Therapeutics, Achira, and A Alpha Bio. 

OpenADMET

OpenADMET has been making waves on several fronts. Their first blog post introducing OpenADMET to the wider community is up, setting a strategic vision to guide the community by developing open datasets and predictive models for ADMET properties. The active learning pipeline is making technical progress, enabling automatic generation of model ensembles across traditional ML and deep learning workflows. Additionally, OpenADMET has deployed its active learning pipeline collaboratively with Octant three times, including a successful compound selection and assay run for a set of nuclear receptors (PXR and AHR). Early iterations have yielded encouraging results and led to concrete model  improvements. 

OpenADMET’s 3D structural modeling pipeline is also making progress, and they have made strides in public data scraping and curation as well as predictive models for physiochemical properties. Collaborative work continues with Octant and UCSF, particularly around compound selection for assays and crystal soaking.  

Behind the scenes, the project has expanded the team's identity and reach, to the point that industry members have reached out to donate ADMET datasets to OpenADMET at OMSF! All donated data will be used to host a blind-challenge over the next two quarters. We’ve also joined Bluesky and LinkedIn, and our blogs are available on Ghost - be sure to follow and subscribe for more news! 

Open Force Field

The OpenFF team has been laser-focused on preparing OpenFF 2.3.0, a new force field incorporating neural network-derived partial charges. The Evaluator package was recently updated to version 0.5.1, resolving a thorny issue that had caused issues with calculations of enthalpies of vaporization, and recent OpenFF Toolkit and Interchange releases added support for NAGL charges in SMIRNOFF force fields, as well as several other improvements.

Meanwhile, Jennifer Clark has continued her groundbreaking work on benchmarking quantum mechanical methods for transition metal complexes. Her project has already produced the largest dataset ever submitted by OpenFF - over 500,000 single-point calculations are complete, with more on the way. This work includes new methodology for estimating the compute requirements of each QM method, laying the groundwork for broader reproducibility and scale.

Community collaboration remains a hallmark of OpenFF. Josh Mitchell has been using Pontibus, a protocol developed by Open Free Energy for calculating solvation free energies using Alchemiscale, developed with major contributions from David Dotson. The OpenFF team has also been cleaning up older data, refining Evaluator, and coordinating with OpenFE and OpenADMET to strengthen inter-project tooling. 

In fact, cross-project collaboration has never been better - Irfan Alibay of OpenFE, along with Matt Thompson, Jeff Wagner, and Chapin Cavender, have been working hard at building out a prototype of SMIRNOFF protein force field handling in Pontibus. This prototype is an important step on the path to eventually letting you use Rosemary in your free energy calculations!

Open Free Energy

This spring saw the release of OpenFE 1.6, which brought significant improvements to the CLI and streamlined workflows for power users. The team has also completed support for OpenMM 8.3, ensuring compatibility with the latest simulation engine builds. Experimental Absolute Binding Free Energy and Separated Topology protocols - complete with notebooks and example systems - are now available through experimental releases, and are being actively validated by the community. The team has also continued to work on the large-scale OpenFE industry benchmark study. We are hoping that our preprint of this work should come out in the next couple of months. In the meantime, you can get a preview via our blog post.

OpenFE has deepened its industry collaborations, particularly with NVIDIA, exploring possible avenues for optimizing our simulations. This includes the use of MPS (Multi-Process Service) to run multiple simulations on a single GPU at a given time. NVIDIA engineers presented their work evaluating the use of MPS with OpenMM and OpenFE at our June board meeting, and we’re grateful for their hands-on engagement. You can read more about this on their blog post. 

The community around OpenFE continues to flourish. Presentations and posters from Alchemistry 2025 and CADD GRC are now archived on Zenodo. If you haven’t already, check out the growing set of talks and tutorials on the OpenFE website.

OpenRosetta

The Rosetta community met at  Summer RosettaCon in August, which was a roaring success! With talks, a workshop, and a hackathon in Seattle, the team saw over 300 community members gather to learn and enjoy the latest in all things Rosetta. Hope Woods joined the team, and she and Rachel Clune have been hard at work cleaning up documentation for RFDiffusion and ProteinMPNN.

This spring, the Foldit team submitted a 273-page proposal for an NIH BTOD Center Grant. A separate Foldit center grant is also in development, in collaboration with the broader Rosetta community. Plans are underway to revise the advanced C++ workshop for Fall, and the FY26 budget has been approved by the board. Cheers to Andreas Petrides for his amazing work!

Looking ahead, multiple events and workshops are planned for this Fall: European RosettaCon is planned for October in Slovenia, AI4PD will happen in November in Chile, and 5 workshops are planned for October / November. You can check them out here: https://rosettacommons.org/education/workshops/

Ecosystem Infrastructure

The EcoInfra team has had a deeply productive quarter. Their work underpins much of the packaging and deployment activity now rolling out across projects. Notably, they’ve taken on OpenFold3’s packaging, including Conda and PyPI distributions, and finalized the internal documentation for our GitHub Actions Runner system, now widely in use by OMSF projects.

One of the major themes of the summer was dependencies. All our software depends on many other packages, which can introduce complications. One complication is that, as new versions of our dependencies are released, it can become difficult to maintain support for both older versions and newer versions. To that end, the Scientific Python community has established a guideline called SPEC0, which provides a timeline for dropping dependency support. The Ecosystem Infrastructure team has developed a Python/command line tool, spec0 to help our teams identify which versions of dependencies they need to support according to that guideline.

Another challenge that arises with dependencies is that they can introduce software vulnerabilities or license conflicts. A software bill of materials (SBOM) is a standardized way to define all the dependencies, and allows for automated license and vulnerability checking. The Ecosystem Infrastructure team has been exploring tooling for SBOMs, to identify tools that will best help our teams keep their code secure and compliant with licenses.

Additionally, they have been developing a tool to easily set up cloud-based temporary development environments. This was required as part of the OpenFold packaging work the team was doing, but had also been identified by the projects as something that could facilitate collaborative projects, provide additional GPUs for developers when their local machines were occupied, or provide GPUs to new hires from day one.

Part of that project is built using infrastructure-as-code (IaC) tools, so the team has also been developing a quick start guide to IaC, aiming to create something as minimal as possible and targeting to OMSF developers, so that any OMSF teams that want to use these tools can get up and running quickly.

They have also developed a set of tools to help our teams test and deploy their project websites. This allows our projects to share common workflows that generate site previews of their websites before deploying to the production site. These tools are now used on all of the OMSF-domain websites (the main website, the OMSF playbooks, and the new software directory listed above), and are also being used by OpenFE and OpenADMET.

Office hours continue monthly, and recent themes have included a “show-and-tell” on how AI tools are supporting your work, and a discussion of project management at the intersection of science and software. However, we’ve started opening our office hours to the public, with our first ever OMSF Office Hours on cloud computing happening in July of last month. We loved seeing the community gather - thanks to Ethan for putting that together!

What’s Next

In the months ahead, expect more releases, better documentation, and continued collaboration across teams. Currently, we are gathering input through upcoming community surveys. If you currently contribute, we would love to have your feedback! Check your inbox!

Thank you, as always, for your contributions to this thriving ecosystem. Whether you’re coding, coordinating, collaborating, or just cheering us on, OMSF is only possible because of you.

Talk to you soon! The OMSF Team