From starting in Math and Computer Science, a PhD in Biochemistry, followed by becoming a business analyst to human genetics research and then clinical trial autism to working on advanced AI technology, to founding a company, Mark DePristo exemplifies real diversity and unbelievable boldness at their best. The driving force, he says, is the willingness to pursue opportunities in adjacent high-growth areas, even if different from what he is presently doing. “Towards the end of every big project, I pop my head up and look for new challenges that will put me at the center of an important technical trend five years in the future. One can know those quite early on and the key is to have the courage to pursue these novel opportunities despite them being new and often outside your comfort zone.”

A builder by background and problem-solver by nature, Mark’s affinity to take very complex details and turn them into understandable stories has helped him discover that AI did not have the same impact in biomedicine as it has in Tech. “Both myself and my co-founder Peyton Greenside, really caught the AI/ML bug towards the end of the 2010s, me coming off four years at Google and Peyton just ending her Schmidt fellowship,” recalls Mark.

But their application domain—biomedicine—wasn’t experiencing the AI revolution that was widespread within tech companies like Google and west coast universities like Stanford. The duo reached a conclusion that AI tech has been so impactful in the tech world precisely because it is integrated directly into the core product stack. A short cycle time between developing a new model, deploying in the real world, and checking whether its predictions are useful in the real world is the driver for why AI technology works so well in Tech. Tech companies leverage these incredibly short feedback loops to quickly improve their algorithms and the product itself. “In contrast, AI technology in Biomedicine is still trapped in a many months longer cycle of large, slow high-throughput experiments.”

These large-scale, slow experiments are valuable for generating training data for machine learning applications in biomedicine. But validating novel predictions of the resulting machine learning models can take months waiting for the next data generation cycle to finish. When the development cycle is measured in months or years, compared to minutes in the Tech world, it is no surprise that AI/ML technology is taking so long to impact biomedicine. By mid-2019, increasingly frustrated by this problem, Mark and Peyton began exploring a better approach to bring AI into biomedicine.

Their initial ideation process focused on identifying a commercially important problem in biomedicine that could be solved by integrating AI/ML tech with a novel type of wet lab enabling fast and small experiments instead of large and slow ones. After a long period of ideation, Mark and Peyton ultimately converged on antibody discovery and engineering as the right domain to develop BigHat’s ML-guided antibody design platform, which uses recent advances in DNA writing and cell free protein expression to rapidly create antibodies and measure their properties. “Previous work, both in the community and from our own research projects, showed that deep learning was revolutionizing protein sequence, structure, and property prediction. Coupling advances in protein modeling with ML to our fast characterization lab would allow us to design, build, and test antibodies in a novel data-driven way. So we founded BigHat in September 2019 and closed our first round eight weeks later,” says Mark.

BigHat’s novel AI-guided experimental platform uses ML techniques ideally suited for multi-parameter optimizations to chart an efficient experimental path to find higher quality variants of an initial antibody. By repeatedly doing rapid design-build-test cycles and using the test data from previous cycles to inform the design for the next cycle, the BigHat platform carries out an efficient, multi-parameter sequence optimization. “This AI-guided smart selection of high-value test sequences enables us to design antibodies with better properties, making them stronger therapeutic candidates, and to tackle antibody designs inaccessible to traditional antibody discovery technologies,” points out Mark.

Mark’s extraordinary skill to take what he sees in the past and use it to figure out something better in the future is the best way to develop a really exciting career. “It keeps you at the forefront of everything, keeps you learning, keeps you challenged, and keeps you from stagnating,” ends Mark.