Bits in Bio Paris: Foundation Models from Proteins to Patients
Come meet old and new friends at the Bits in Bio Meetup, happening on September 22nd from 5:30 PM to 7:30 PM at the InstaDeep Paris office. This edition brings together three teams InstaDeep, Orakl Oncology and Scienta Lab around a shared question: how far can learned representations of proteins, cells and patients actually take us, and where do those models still hit a wall?
Bits in Bio is a global community building at the intersection of software and biotech, with the mission to bring together tech and bio communities to share knowledge, build software tools, and empower scientific discovery. We organize meetups, Q&As with industry-leading figures, hackathons, and more to facilitate connections in this space.
Program:
6:30 – 7:00 PM: Welcome & Registration
7:00 – 8:15 PM: Presentations (25 min each, including Q&A)
PRIMA: a bidirectional state-space architecture and training approach for sequence modelling of protein-protein interactions by Arturo Fiorellini-Bernardis, Senior Research Engineer @InstaDeep An alternative to transformer-based protein language models that scales linearly with sequence length and beats them at a fraction of the compute.
EVA-RNA: learning patient representations from transcriptomics at scale by Yannis Cattan, AI Research Scientist @Scienta Lab A 300M-parameter encoder pre-trained on 500k+ samples, built to represent patients in a learnt latent space.
Noise ceilings and modular benchmarking for omics-based cancer drug response prediction by Lélia Polit, Head of Computational Biology @Orakl Oncology A long-standing ML problem and a decade of increasingly complex architectures. Measured against a replicate-derived noise ceiling, how far have we actually come?
8:15 – 9:00 PM: Networking
Join us for an evening of insightful talks, networking, and collaboration with peers at the forefront of biotech and AI!
Photos will be taken during the event and may be used on InstaDeep and Bits in Bio channels. Just let a member of the team know when you arrive if you'd rather not appear.