Valence Labs

The Valence team was proud to share our research at #NeurIPS2024 last week. We had 14 papers across the main conference and workshops. At the Foundation Models for Science workshop, Valence and Recursion researchers won first, second, and third place for our work on: “How Molecules Impact Cells: Unlocking Contrastive PhenoMolecular Retrieval” Predicting the effect of molecules on cells with MolPhenix Paper: https://lnkd.in/ekeMTSzj Dominique Beaini ”GFlowNet Pretraining with Inexpensive Rewards” Atomic GFlowNets for de novo molecular design Paper: https://lnkd.in/egHxT4Q9 Mohit Pandey “ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy” Our largest foundation model for cell microscopy to date Paper: https://lnkd.in/eb7wFmMz Kian Kenyon-Dean Looking ahead to ICLR 2025, members of the Valence team will be part of the organizing committee for the Learning Meaningful Representations of Life (LMRL) workshop. LMRL welcomes submissions on a wide range of topics, including but not limited to: multimodal representation learning, multiscale representation learning to connect molecular and biological data, modeling biological perturbations, and more. More details: https://www.lmrl.org/ We are also hiring for multiple roles across our offices in Montreal and London. If you’re interested in helping advance our mission of industrializing scientific discovery to radically improve lives, join us: https://lnkd.in/g9h5beuK

Valence scientist Michael Craig will serve as the Senior Chair for the Drug Discovery and Development roundtable at the ML4H conference, taking place at 12:30 PM today. This session will focus on the role of Large Language Models and Foundation Models in accelerating drug discovery. It will address key challenges in leveraging these models to optimize the drug development pipeline. For more information on the roundtables, visit: https://lnkd.in/g3B9PB_Y

Definitely a career highlight to be awarded a best paper award at NeurIPS FM4Science 🤩🤩 But even more exciting is that the 2nd and 3rd awards are given to colleagues Kian Kenyon-Dean and Mohit Pandey, a podium sweep from Valence Labs / Recursion 🥇🥈🥉

Thank you to everyone who attended our event at #NeurIPS2024 with Recursion and NVIDIA earlier this week. If you didn’t get a chance to talk to our team, we have multiple posters and orals at the workshops today. We’re also hiring for multiple full-time and internship positions. Explore our open roles: https://lnkd.in/g9h5beuK Here’s where you can find us: “Benchmarking Transcriptomics Foundation Models for Perturbation Analysis” Where: AIDrugX workshop in West Meeting Room 109, 110 https://lnkd.in/gsDCEsvu Alisandra Denton “Score-Based Interaction Testing in Pairwise Experiments” Where: Causal representation learning workshop in East Exhibition Hall C https://lnkd.in/g_E2cwaP Jason Hartford “Towards Scientific Discovery with Dictionary Learning: Extracting Biological Concepts from Microscopy Foundation Models” Where: Interpretable AI workshop in East Ballroom A, B https://lnkd.in/gQ369xAR Jason Hartford “ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy” Where: Foundation Models for Science workshop in West Meeting Room 202-204 https://lnkd.in/gmKtVVxP Jason Hartford Oren Kraus "SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints" Where: Machine Learning in Structural Biology workshop in East Meeting Room Rooms 11 & 12 https://lnkd.in/gQPsCH_s Miruna Cretu Emmanuel Bengio

While generative models have great potential to accelerate drug design, they often produce molecules are that hard to synthesize. At #NeurIPS2024, we’ll share more about SynFlowNet, a GFlowNet specifically trained to generate molecules from documented chemical reactions and purchasable starting materials. This lets us constrain the exploration of such a generative model to a synthetically accessible chemical space and sample not only target compounds but also the synthetic routes that lead to them. Find us at the MLSB workshop on Dec 15th at 11 AM PT in the East meeting rooms 11 & 12. https://lnkd.in/gWVMpfDa

Last NeurIPS, we introduced Sequential Attachment-based Fragment Embeddings (SAFE), a novel line notation for chemical structures. SAFE reimagines SMILES strings as an unordered sequence of interconnected fragment blocks while maintaining compatibility with existing SMILES parsers. It streamlines complex generative tasks, including scaffold decoration, fragment linking, polymer generation, and scaffold hopping, while facilitating autoregressive generation for fragment-constrained design. This eliminates the need for intricate decoding or graph-based models. The team behind SAFE will be presenting a follow-up to this work: “SAFE setup for generative molecular design” at the AI4Mat workshop tomorrow at 12 PM PST. Come and meet the team in the West Meeting Room (211-214). https://lnkd.in/ga8EHwSc Read the paper: https://lnkd.in/g8KSmv3S

Missed the event yesterday? Our team is still at NeurIPS today and throughout the workshops. Come talk to us and learn more about open roles. We’re hiring for both full-time and internship positions: https://lnkd.in/g9h5beuK

Tonight is our TechBio social with Recursion and NVIDIA. Meet fellow scientists, engineers, and founders working at the intersection of ML and drug discovery at Science World. Space is limited. Join the waitlist: https://lu.ma/biikt7ox

How do we enable a multimodal future in TechBio? Come chat with us at NeurIPS today to learn more. We’ll be at East Exhibit Hall A-C #1000 from 11 a.m. to 2 p.m. PST, presenting our paper "Propensity Score Alignment of Unpaired Multimodal Data." Multimodal representation learning techniques typically rely on paired samples to learn common representations, but paired samples are challenging to collect in fields such as biology where measurement devices often destroy the samples. In our paper “Propensity Score Alignment of Unpaired Multimodal Data” we explore alignment techniques that allow us to estimate a common space in which to match samples. In this case, we match phenomics and transcriptomics data together. Quanhan (Johnny) Xi Jana Osea Zuheng (David) Xu Jason Hartford https://lnkd.in/gwcE2VND Read the paper: https://lnkd.in/gv-fErT5

Phenomics refers to the systematic study of a cell's phenotype—encompassing its observable and functional characteristics—in response to various perturbations. Leveraging the capabilities of Recursion’s automated labs, we have access to billions of high-resolution cell images, each capturing intricate details of cellular morphology, including shape, size, texture, and subcellular structures. Using a small subset of this data, we created MolPhenix: a foundation model that can predict the effect of any given molecule and concentration pair on phenotypic cell assays and cell morphology. We’ll be presenting MolPhenix at NeurIPS this year with our poster “How Molecules Impact Cells: Unlocking Contrastive PhenoMolecular Retrieval”. Come chat with us next week on Dec 11th from 11 - 2 PM PST at the East Exhibit Hall A-C #1110. Read the blog: https://lnkd.in/eWpuH9aq Read the paper: https://lnkd.in/epDf7Md3