Exciting and inspiring- those are the best words to describe the OpenFold consortium meetings and discussions. Haven’t heard of OpenFold? You wouldn’t be alone, but you’ve probably heard of AlphaFold2.
In the drug discovery world, AlphaFold2 is now almost synonymous with solving one of the most elusive problems in computational biology and chemistry: protein folding. AlphaFold2, as the name implies, is built upon its predecessor AlphaFold’s success at the 2021 14th Critical Assessment of Structure Prediction (CASP14) where it trounced the competition by accurately predicting the unknown structures of proteins from their amino acid sequence. Developed by DeepMind, a subsidiary of Alphabet Inc., AlphaFold2 utilizes a deep neural network architecture trained on a vast dataset of known protein structures to predict the complex folding patterns of proteins with ground-breaking accuracy. In a surprisingly bold move, DeepMind departed from the conventional practice of keeping such cutting-edge algorithms proprietary and published comprehensive details of its methodology and findings in July 2022. But that move may be seen in a different light when you learn a little more about OpenFold.
AI for All: OpenFold’s Role in Drug Discovery
OpenFold was founded in February 2022 by the AlQuraishi lab at Columbia University, Arzeda, Cyrus Biotechnology, Outpace Bio and Genentech’s Prescient Design. OpenFold is modeled after pre-competitive technology industry consortia, embodying the ethos of open science, encouraging transparency, shared knowledge, and accelerating scientific breakthroughs. Its first release in June 2022 included not just the inference code and model parameters reproducing and improving upon AlphaFold2’s speed and accuracy, but also full training code that could allow a full set of derivative models to be trained for specialized uses in drug discovery of biologics, small molecules, and other modalities.
As stated on its web pages, OpenFold “is a non-profit AI research and development consortium developing free and open-source software tools for biology and drug discovery”. It aims to democratize the power of biological AI and lead a community-driven effort to make advanced protein structure prediction tools available to researchers and scientists in academia, biotech and pharmaceutical companies across the globe. Without a doubt, its inception played a part in encouraging DeepMind to share the details of their discovery and has ushered an exciting new era of transparent and collaborative research. DeepMind have continued their development of AlphaFold models. In October 2023, they shared the initial details of their work to expand coverage beyond proteins to other biological molecules, including ligands. This effort may help crack long-standing challenges such as accurately predicting protein-ligand structures and potentially replacing current industry standard docking methods. This work was done with Isomorphic Labs, a commercial venture stemming from the original AlphaFold DeepMind team, who are reimagining the “entire drug discovery process from first principles with an AI-first approach”. The work of OpenFold and other academic leaders such as Baker’s lab will hopefully incentivize the continued sharing of knowledge to promote advancements in bioinformatics, drug discovery, and structural biology that wouldn’t be imaginable five years ago.
Embracing a New Era: OpenFold’s Journey to 3DEXPERIENCE Platform
BIOVIA Dassault Systèmes became an industry consortium member of OpenFold in November 2022, being drawn to OpenFold’s mission to develop an open ecosystem of accelerated AI-based tools for biology and drug discovery research. Our impetus is to deliver valuable software solutions with the best methodologies for pharmaceutical and agricultural product design, helping our users solve real-world problems in drug discovery, disease understanding, and bioengineering. We are now enriching our drug design experiences by amplifying the power of long-established and validated physics-based modeling and simulation methods with cutting-edge AI methods. In February 2024, the OpenFold (monomer) and the AlphaFold (multimer) models will be available to BIOVIA Discovery Studio Simulation users on the 3DEPXERIENCE platform as an alternative to the traditional homology modeling algorithm MODELER. Subsequent advances will assimilate the AI models into further simulation workflows addressing other challenges that our users have and look towards the burgeoning interest in generative biological design.
Excitement and inspiration are what I take away from each of the OpenFold consortium meetings. In this time of dizzyingly breakneck developments in AI, its communal space for data sharing, discussions, and improvements really allows me to keep abreast of the development of new tools, models and benchmarks and deliver what our industry-leading customers need in the ever-changing landscape of drug discovery.