Tristan Bepler

Chief Executive Officer
​NE47 Bio Inc

​ Tristan Bepler is the Group Leader of the Simons Machine Learning Center at the New York Structural Biology Center in NY, USA and the co-founder and CEO of OpenProtein.AI. He received a BS in Computer Science and Biology from Duke University and a PhD in Computational and Systems Biology from MIT where he was advised by Bonnie Berger. During his PhD, Tristan pioneered deep protein language models for learning protein representations from large natural protein sequence databases and leveraging these representations to elucidate the protein sequence-structure-function relationship. Tristan continues to work on machine learning on proteins, especially methods for data efficient, function-based protein engineering. He also develops machine learning methods for understanding protein structures and dynamics using cryo-electron microscopy. His group develops and maintains several software packages (Topaz, Ptlolemy, CryoDRGN) that are widely used by the cryoEM structural biology community. Tristan’s work has been cited hundreds of times and published in top biology and machine learning venues.

Speech title & Synopsis

Machine Learning Driven Protein Engineering

Over the past few years, machine learning has enabled huge advances in our ability to understand and predict protein properties. Large protein language models are able to learn from evolution to generate functional proteins de novo and accurately predict variant effects. AlphaFold2 and RoseTTAFold have revolutionized protein structure prediction. New versions, extensions, and applications of these machine learning models are being explored every day. In this talk, I will discuss recent advances in protein machine learning with a focus on their applications to protein engineering. I will show that machine learning can be used to design protein variants with significantly improved efficacy, as demonstrated by a case study in designing potent SARS-CoV-2 binding antibodies. Despite their promise, many challenges remain in incorporating these very large models into routine protein engineering methods. Our OpenProtein.AI platform aims to address this by providing access to state-of-the-art algorithms through intuitive and user-friendly workflows in a web-based system.
​​