Apheris

The last few weeks were among the busiest we've ever had. We published the first results from AISB-1, launched AISB Bind, announced the founding members of the Antibody Developability Consortium with Ginkgo Datapoints, and completed Run 3 of the ADMET Network. My October newsletter covers my take on all of it, plus what to expect at our evening with SandboxAQ in Berlin on October 19. None of this happens without the teams at our partner companies, and without the Apheris team. Thank you all! Frank Oellien, John Karanicolas, Maria Kadukova, Lucian Chan, Carl Poelking, Paul Mortenson, Chris Murray, Matt Pokross, Payal Sheth, Jose Carlos Gómez Tamayo, Gary Tresadern, Edward King, Prashanth Vishwanath, Anders Hogner, Justin Scheer, John Androsavich, Ena Cratsenburg, Rich Cohen, Joshua Moller Athena Hadjixenofontos, PhD, Bryan Rutsch, Erwin Pannecoucke, Henrik Keränen, Allan Jensen, Yves Fomekong Nanfack, Charlotte Deane, Peter Tessier, Lukas Jarosch, Vinay Swamy, Mohammed AlQuraishi, Andrew McLaughlin, Bill Fitzgerald, Jennifer Stewart, MBA and last but not least the whole Apheris team 👏 #FederatedLearning #DrugDiscovery #CoFolding

The night before NVIDIA GTC Berlin, we're hosting an evening for the AI, healthtech and life sciences community, and you're invited. Apheris and SandboxAQ are teaming up for an evening in Berlin with the people working on AI in science. You'll hear from SandboxAQ's COO Andrew McLaughlin, Apheris' CEO Robin Röhm, and the researchers behind a new model and network partnership we'll announce that night. SandboxAQ builds Large Quantitative Models, grounded in physics and chemistry, that reach from catalyst discovery (AQCat) to a virtual cell. We look forward to many groundbreaking insights and then: food, drinks, music, and time to chat. 📍 Where: zazou, a beautifully restored 1881 bank building in central Berlin 📅 When: Monday 19 October, the evening before GTC, starting at 7pm Spots are limited. Register to join us 👉 https://luma.com/8mw452pu #NVIDIAGTC #AI #DrugDiscovery #LifeSciences #HealthTech

Single-point % inhibition is abundant, but most affinity heads can't consume it. The right one can.

Ternary complexes like molecular glues and PROTACs need models that see more than one chain.

Fine-tuning an affinity head on only ~50 compounds can restore series ranking in under an hour.

A model that can't tell weak binders from strong ones can't prioritize what to synthesize next.