Axial Podcast

Hannes is a graduate student at MIT working towards a PhD in computer science. Within the span of a few months, he has been able to publish two ICML papers: one focused on molecular property prediction and the other developing a model, called EquiBind, for drug binding prediction. Hannes, advised by both Regina Barzilay and Tommi Jaakkola, is doing exciting research at the intersection of graph machine learning and biology. In this conversation, we discuss his career so far starting at Munich to an internship at MIT that ultimately led to Hannes starting his PhD at MIT earlier this summer. For Hannes, biology is a new frontier to not only develop new ML models but have an impact on human health. We get into the overall field of molecular dynamics and machine learning. Everything from the need for higher quality datasets in life sciences to new models. We touch upon the advantages of graph models to predict biological phenomena and the opportunity to feed these models better data. Afterwards, we go into his two papers - InfoMax and EquiBind. The former uses pre-training to capture implicit 3D structures of small molecules to predict various properties. And the latter relies on graph models to avoid thorough sampling thereby speeding up the process of predicting ligand/target binding. Both of these methods are interesting case studies on the power of graphs and the potential of relying on new models and data to map out biology rather than sheer computing power. Finally, we talk about longer-term opportunities in ML and self-supervised learning. Everything from protein to small molecule design. For Hannes, modeling molecular interactions is the key focus he takes and he's spending more time thinking about proteins and their activity in cells. I would also recommend joining the reading group he leads on graph machine learning. From our conversation, a favorite quote of mine was Hannes focus on doing work with "an eye for good."

We talk with Justin Farlow, Co-Founder and CTO of Serotiny about his journey from UCSF to founding a company with his brother, Colin. In this conversation, Justin discusses his initial discovery of engineer-able biology from a physics lens to earning his PhD at UCSF under Zev Gartner while being in the epicenter of both synthetic biology and software startups. Then he goes into his journey as a founder, starting Serotiny initially as a SaaS company then pivoting toward building a wet-lab platform after the approval of the first CAR T therapies. Mammalian synthetic biology promised curative therapies in both new cell and gene therapies, and the rapid progress of these new modalities helped Serotiny build a unique business model exemplified by recent deals with both Janssen and Tessera Therapeutics. With more likely in the pipeline. Serotiny is the market leader for designing therapeutic multi-domain proteins - from chimeric antigen receptors (CAR) to CRISPR gene editors, where the aim of the protein is to change the properties of a cell. Their platform relies on machine-guided variation to design in silico libraries of millions of protein designs and then test tens of thousands of them in vitro, and iterate to produce a high-value candidate. Versus the past state-of-the-art, Serotiny enables unbiased screening of large protein therapeutics in their native mammalian and therapeutic contexts. Unbiased screening for complex drugs has allowed the company to find new candidate combinations that are hard-if-not-impossible to discover with other approaches. By generating and intentionally structuring data that correlates primary amino acid with primary cell phenotype, the company's underlying platform is allowing Serotiny to move more quickly from idea to drug candidate. At the end of the conversation, we discuss the long-term need for a common language in synthetic biology, building a world-class team, and the opportunities to standardize datasets in life sciences. Justin lays out a powerful framework for platform companies in drug development: going 0 to 1 to find a signal and invent a new candidate and then going from 1 to 100 and beyond by versioning the candidate to improve its therapeutic potential.

King is currently a graduate student in the Howard Chang Lab at Stanford. In this conversation, we discuss everything from the beginning of his scientific career to extrachromosomal DNA (ecDNA) and cancer drug development. King went to college at the University of Washington where he became passionate about developmental biology. The beauty of a FISH experiment hooked King to lab work and set him on a path to become a world-class biologist. He chose to go to graduate school at Stanford and join the Chang Lab to bring together various genomic tools to study cancer development. King was recently the lead author on a Nature paper discovering ecDNA hubs and establishing a set of rules on how these hubs increase oncogene expression. Extrachromosomal DNA is thought to increase tumor proliferation through amplification and increases in expression of oncogenes. King's work found that there are clusters of 10-100 ecDNA hubs within the nucleus that promote oncogene overexpression. Within these hubs, an individual ecDNA is more likely to express an oncogene when it spatially clusters with other ecDNAs. The paper combines chemical perturbation to verify tethers for these hubs, CRISPRi screening to map out enhancer-gene activation relationships within an ecDNA, high-resolution imaging, and 3D genome tools. Truly a tour de force. This work sets the table for an entirely new class of druggable targets and pathways for cancer development. At the end of the conversation, we discuss the potential to drug ecDNA hubs and King's future work in the field. It's obvious from his research and his commitment to excellence, King is going to continue putting out groundbreaking research in genomics, cancer, and more.

Really excited to put out this conversation with Kadi Saar, an outstanding inventor and world-class computational chemist, where we discuss her career spanning childhood interests to graduate research to her work now around high-throughput structural biology and condensates. Kadi is a rising star in life sciences building out new experimental and computational tools to probe biology and advance human health. Early on in her life, it was pretty obvious that Kadi was very unique. She won Estonia's Got Talent for mental arithmetics. You can watch her talent here: https://www.youtube.com/watch?v=wBNrq28ADQY I give Kadi a set of numbers to multiply to see if she still has it. She was a world-ranked tennis player growing up, and ultimately had to make the tough decision to focus on chemistry and engineering upon entering university. She went to Cambridge (Trinity College) studying chemical engineering and biotechnology and conducted her PhD research under Tuomas Knowles also at the University of Cambridge. Afterwards, Kadi did a postdoc as a Schmidt Science Fellow under Knowles and Alpha Lee. While doing her postdoc, Kadi published really exciting research around merging high-throughput structural biology and computational chemistry to predict protein:ligand affinities and improve virtual screening for drug development. I would really recommend reading her preprint and we also touch upon it later in the conversation: https://www.biorxiv.org/content/10.1101/2021.10.15.464568v1.full.pdf In short, the preprint shows that ~100 diverse structures of a ligand in complex with a target protein, using a COVID-19 protease as a proof-of-concept, along with first comparing compounds pairwise before ranking them by bioactivity to remove experimental noise is sufficient to virtually design high affinity binders. Kadi, in my opinion, is a great example to follow. The conversation goes a lot into her journey towards becoming a scientist and will serve as a resource for years to come. Looking forward to doing another one in a few years and getting an update on Kadi's work. A favorite quote of mine after speaking with Kadi is working on projects where she "likes the people", in particular when having to choose between multiple equally exciting directions.

We discuss Bruce's career from being pre-med and getting into research while at Washington University in St. Louis to working at Intrexon and going to grad school at Caltech afterwards. While at Intrexon, he saw the need for better tools to scale biology and decided to go to Caltech to pursue his ideas. At Caltech, he was advised by Prof. Frances Arnold, who pioneered the protein engineering technique "directed evolution" that eventually led to her winning the Nobel Prize in Chemistry in 2018. While a member of the Arnold Lab, Bruce was part of a group bringing machine learning to protein engineering and directed evolution; over the course of 5 years, Bruce did some incredible work in grad school. A key paper was published in Cell around machine learning-assisted directed protein evolution: https://www.cell.com/cell-systems/fulltext/S2405-4712(21)00286-6 You can also read his graduate dissertation here: https://thesis.library.caltech.edu/14631/ Beyond his research and numerous papers, we also discuss the broader field of protein engineering and ML and his perspective on new opportunities in comp bio and protein design. On top of all of this great work, Bruce along with several people from the Arnold Lab maintain one of the best documented GitHub repos in bio: https://github.com/fhalab/MLDE. A favorite quote of mine after speaking with Bruce is around his goal to build tools that "are accessible to as many people as possible."

We discuss Nader's career from learning structural biology in the Marqusee and Alber Labs at UC Berkeley to his pioneering graduate work at MIT with Forest White and Li-Huei Tsai and now at Stevens at Harvard Medical School. The conversation is centered around the 2 groundbreaking papers he published while earning his PhD: 1. Phosphoproteomics identifies microglial Siglec-F inflammatory response during neurodegeneration 2. Quantitative phosphoproteomics uncovers dysregulated kinase networks in Alzheimer’s disease His research uses phosphoproteomics to capture the heterogeneity in Alzheimer's disease progression and uncover new leads to understand neurodegeneration. Genetic studies like GWAS will only find risk factors so proteomics and other tools are needed to comprehensively understand new biology and disease. Nader's work is centered around new models, tools, and targets. In particular, we need better mouse models to capture the same genes and capture human pathology. For example, Nader's work has found glial cell protein clusters as a promising lead that increase in expression slightly later than tau but earlier than neurodegeneration. This observation would not have been captured by a purely genomics approach. I am sure the best is yet to come given the impact of his research. A favorite quote of mine after speaking with Nader is around his long-term goal to "inspire the next-generation of scientists."

We discuss Diego's career from building GeneWeave (acquired by Roche) to Endpoint Health now. Endpoint is pioneering a new business model merging diagnostics, data, and drug development to start with patient data and back into a therapy. The company is using their platform to develop first-line medicines for indications in infectious diseases with no approved therapies and beyond. Truly, Endpoint is leading a new wave of progress within critical and chronic illnesses. A favorite quote of mine after speaking with Diego: during an acquisition, "as an entrepreneur, you do a deal like this once." so while at GeneWeave they hired a bank to organize the entire process.

We discuss James' and Deniz's careers from how they first met to founding Seven Bridges and Totient and now building AbSci after Totient was recently acquired. With experience leading teams in the 100s spanning genomics and SaaS to drug development and even paint from James' childhood entrepreneurial pursuits, both have an incredible amount of wisdom on scaling technology companies. We touch on AbSci's unique model of accelerating and lowering the barriers for biologics development and talk about the finding the right platform-partner fit. Some favorite quotes of mine after this conversation are: during any negotiations, "competition is highly important for deals" and Deniz's experience of "seeing the future early but unevenly distributed" just as sequencing costs began coming down outpacing Moore's Law.

We discuss Jacob's career going from physics to biology and making the transition from academia to industry. Building and leading data science teams at GNS Healthcare, Indigo, and now EQRx, Jacob is one of the best data scientists in biotech. In our conversation, we talk about ways to generate standardized data for machine learning models, building interdisciplinary teams, and implementing relevant models for drug development.  Then we touch on EQRx's fast follower drug development model and the role of data in integrating decision making across the board from target ID to commercialization. A favorite quote of mine from talking to Jacob was: "things don't change much until they change all at once."

We discuss Brian's career starting at Trinity in Dublin to Stanford, 23andMe, and Hexagon. In our conversation, we talk about the interplay between computation and biology, business models in biotech, and what it takes to build world-class teams. Brian really has the superpower of building and being part of talent hubs. If he ever wants to, I know he would build the best talent agency in biotech.