Making Big Moves and Large Molecules

For the past 50 years, structural biologists, biochemists, and computational scientists have been on the hunt to develop more efficient and accurate ways to predict how proteins fold into their three-dimensional structure solely from their amino acid sequence. A protein’s structure, which is determined from a linear sequence of amino acids, dictates its function in a cell but when misfolded or missing has the potential to contribute to disease. The development of machine learning (ML) methods like AlphaFold2 and RosettaFold that predict protein-structure have recently led to major advances in solving the problem, paving the way for the field to now address a different, yet related, challenge using ML: protein-ligand and protein-protein interactions, central to understanding disease biology and drug discovery.

Antibodies, now one of the most common and powerful treatment modalities for many diseases, and the antigens they bind on a tumor cell or in diseased tissue, are both proteins. How these two proteins interact dictates how well certain antibodies work in treating disease. To further extend ML developments, Genentech seeks to use ML to enhance antibody discovery. “Using machine learning to be able to identify, optimize and design antibodies will not only allow us to make better lead molecules but also allow us to make them more efficiently. Which I’m hoping will translate to us bringing antibody therapies to clinical trials and potentially patients, more quickly.” says Arvind Rajpal, vice president, Large Molecule Drug Discovery, Genentech.

This goal led to the recent Genentech acquisition of Prescient Design, a company originating in New York City and focused on applying ML to protein design.

“We have developed a unique program known as the manifold sampler, which allows a scientist to make guided changes to a protein sequence, saving a lot of time when designing a protein such as an antibody.” explains Richard Bonneau, co-founder and executive director, Prescient Design, Genentech.

With potentially the right ML model to start retraining and modifying for protein-protein interactions, Prescient Design was of interest to Genentech and, as Rajpal recalls, the other big factor was the talent of the founders who bring a “multifactorial perspective to the difficult problem of predicting antigen-antibody interaction.”

The Prescient Design team joined Genentech in August 2021 as a new department, giving them the autonomy and freedom to continue to build their team and technology, while also allowing them to seamlessly integrate their expertise in ML into Genentech’s existing research and drug development. Their “lab in the loop” model allows a harmonious flow of data and information from experimental scientists back to computational teams that helps to refine models which can then be used to inform new experiments.

“When Genentech approached us, we were really happy to gain significantly more opportunities to test our methods and platform. We can work with Genentech’s best in class R&D labs, we'll have a team of people to further develop actual molecules that we identify, and we get to focus on things that we know best,” says Vladimir Gligorijevic, co-founder and senior director of AI and ML, Prescient Design, Genentech.

While the Prescient Design platform will initially support Genentech’s antibody discovery efforts, the technology has the potential to accelerate numerous aspects of drug discovery and help address challenges beyond protein-ligand and protein-protein interactions. With the convergence of biology and advanced computation, including ML, this acquisition is an exciting element of Genentech and Roche’s broader strategy to use computational approaches and predictive models to boost and enhance the current drug discovery and development process, with the goal of delivering better medicines to patients faster. “By bringing further advances in machine learning and artificial intelligence to Genentech, my goal is to help make the organization future-proof and develop potential medicines that treat diseases we haven’t been able to tackle,” says Kyunghyun Cho, Associate Professor of Computer Science and Data Science, NYU and co-founder and senior director of Prescient Design, Genentech.