Bristol Myers Squibb, Takeda, and Astex Pharma have joined a recently formed consortium that is working on refining the OpenFold structure prediction platform with pharma industry data.
The AI Structural Biology (AISB) consortium was set up earlier this year to fine-tune OpenFold3, which is being developed by Mohammed AlQuraishi’s team at Columbia University, with AbbVie and Johnson & Johnson the first drugmakers signing up to the project.
The main thrust of the project, which is supported by secure network specialist Apheris, is to overcome one of the main challenges facing the developers of AIs for predicting molecular structure – the limited availability of data on protein and ligand structures necessary for training the models to make sure they are useful for drug developers.
By necessity, models of this kind are being built using flawed, publicly available databases that lack the precision to deliver the predictive accuracy and generalisability needed for complex drug discovery, and layering in pharma industry data can provide big improvements.
It is viewed as an alternative to Google DeepMind and Isomorphic Labs’ AlphaFold project, but has been built with drug development as the primary objective.
The five pharma partners have agreed to open up their proprietary, structurally derived data to train the system within a secure environment designed to protect their trade secrets and intellectual property.
They are all contributing data from several thousand experimentally determined protein–small molecule structures, creating what Apheris has called “one of the most diverse datasets assembled for model training in drug discovery.”
“Our AI and machine learning strategy is deeply embedded in our R&D framework, guiding everything from target prioritisation to predictive molecule design and patient segmentation,” commented Payal Sheth, head of discovery biotherapeutics and lead discovery and optimisation at BMS.
“We’re bringing together diverse structural datasets from multiple pharmaceutical companies to advance predictive models for small molecule discovery in ways no single organisation could achieve alone,” she added.
Ultimately, the goal is to make OpenFold3 a tool to accelerate molecular design by achieving predictive precision comparable to experimental methods like X-ray crystallography, according to Apheris.
The AISB is also looking into taking a similar, federated approach to additional initiatives in small and large molecules.
“In drug discovery, no single company has enough data to solve the hardest problems alone,” said Robin Röhm, co-founder and chief executive of Apheris.
“Federated learning allows the industry to overcome this barrier,” he added. “With the AISB network, we’re showing that it is possible to combine the power of pharma datasets without ever moving or exposing them.”
