AI in Life Sciences: Protein Representation and Autonomous Labs

Tokyo AI hosts an evening on the intersection of applied machine learning, computational biology, and wet-lab automation. Three talks walk up the stack: representation learning that places molecules and proteins in a shared vector space, genome-scale language modeling over a 2 million base pair context window, and agentic platforms that turn fragmented scientific literature into actionable experimental layouts. Stefano Massaroli (co-founder, Radical Numerics) opens with Omnii, a genome language model pretrained with native fusion of annotation tracks alongside DNA sequence, covering block convolutions, sparse attention, and the benchmark picture on clinical variant interpretation. Daniel Leuck (Ikayzo) and Calvin Duong (University of Tokyo) follow with Reveria, a beta platform of modular CodeLets for AI-assisted wet-lab work, using influenza research as the case study. Ewa Szyszka (Qdrant) closes on DrugCLIP, which recasts "which compound binds this target?" as a vector search problem over ESM-2 protein embeddings. The audience is researchers, engineers, and technical product managers in drug discovery and applied ML. Doors open at 18:00, talks run 18:30 to 20:00, and networking follows until 21:00. The through-line is a practical one: screening happens virtually, while physical wet-lab validation remains the core bottleneck.