We scan new podcasts and send you the top 5 insights daily.
Younger says de novo designs currently achieve reasonable single-digit percent success rates only against the easiest targets and epitopes. The field's ambition is to engineer challenging epitopes and additional properties, where success rates remain very low; he argues more and better data are still needed.
To break the data bottleneck in AI protein engineering, companies now generate massive synthetic datasets. By creating novel "synthetic epitopes" and measuring their binding, they can produce thousands of validated positive and negative training examples in a single experiment, massively accelerating model development.
De novo design is not a magic bullet, but it's a powerful new tool. Major pharmaceutical companies report it successfully generates binders for difficult targets where conventional methods like immunization have failed, effectively closing critical gaps in the discovery pipeline.
Despite widespread industry hype around in silico antibody generation, screening exposed human B cells captures natural structural diversity that algorithms miss. Platforms focusing on human biology rely on evolutionary optimization to uncover functional antibodies for hard-to-drug targets.
Drug developers can now combine multiple specificities into hyper-complex molecules. However, there's concern that this engineering-first approach, without a strong biological hypothesis, creates constructs too complex to predict in vivo, risking a return to less-targeted screening methods.
Younger notes the PDB shows how proteins physically interact but not how strongly, and it has almost no examples of antibodies one or two mutations apart with measured affinities. Zero-shot optimization instead needs local affinity landscapes built from many single-point mutation variants across many targets.
Instead of screening billions of nature's existing proteins (a search problem), AI-powered de novo design creates entirely new proteins for specific functions from scratch. This moves the paradigm from hoping to find a match to intentionally engineering the desired molecule.
De novo AI is proving its value against notoriously difficult targets. Panelists from major pharmaceutical companies confirmed that these methods are achieving early, promising successes against targets like GPCRs, which have historically been challenging for conventional antibody discovery platforms.
Dr. Crowe believes AI's most immediate impact isn't designing new antibodies from scratch. Instead, it's analyzing vast datasets to predict which newly discovered antibodies are developable (stable, high-expression), preventing investment in potent molecules that ultimately can't be manufactured at scale.
David Younger says his company's model training showed that solving truly hard protein design problems will require much more data than the field expects. He also warns public binding and affinity data are often not interoperable, lack proper controls, and lack annotations, making them hard to reuse.
Younger says the Protein Data Bank today holds about 10,000 antibody-antigen structures and adds roughly 1,000 yearly. Their internal scaling work suggests good generalizable de novo design models may need hundreds of thousands to millions of structures, so waiting on PDB growth is far too slow.