Democratizing AI Innovation for Protein Engineering

Recent advances in Deep Learning have enabled new application domains in molecular biology and drug discovery. The triple exponential growth due to increasing abundance of data, improving compute resources and novel algorithms allows to tackle problems considered unsolvable only few years prior, such as protein folding problem. Such approaches are currently available to select few, who have access to necessary resources, and sufficient know-how to benefit from such methods.

This talk will discuss real life examples of research problems, that modern AI methods could address, including answering unasked scientific questions and provide unforeseen experimental leads. We will also discuss AI-aided methods for in-silico hypothesis validation and opportunities for rapid iterative discovery cycles.

Large ML models trained on extensive data corpora allow for constructing highly accurate, tailored models proposing bias-free, actionable hypotheses. Training such models and enabling unrestrained access to them will enable end-users to apply them to own research. These applications in turn will feed back to community, permitting for better, more applicable ML methods becoming available. None of it is possible without use of large scale HPC systems in a scalable, replicable way.