Exploring the Power of Foundation Models at the Tahoe Therapeutics Hackathon

Last week, I had the incredible opportunity to participate in a fast-paced and inspiring hackathon organized by Tahoe Therapeutics (formerly Vevo). We have focused on Foundation models, single-cell transcriptomics, large scale data processing and their applications in drug discovery. Over just 1.5 days, our small team of five scientists came together to explore how large-scale gene expression data can drive innovation in computational biology.

Building, Benchmarking, Discovering

Together with Sreenath Srikrishnan, Constantin Goeldel, Antoine Argante, and Meriem Bensouda, Ph.D., we tackled a challenge that sits at the heart of drug discovery: predicting gene expression responses to perturbations. Using the Tahoe-100M dataset, a rich resource of single-cell data under diverse conditions, we set out to benchmark both foundational models (Transcriptformer, ContrastiveVI, scVI) and simpler baselines (PCA) on their ability to capture complex transcriptional changes.

It was impressive to see what could be accomplished by a small, focused team with the right mix of data, compute power, and creative energy. In just over a day, we designed experiments, built models, ran evaluations, and generated insights that could serve as a stepping stone for future research and applications.

Why It Matters

The Tahoe-100M dataset is a potential gold mine for computational biology. Its scale and diversity open doors to a wide range of applications — from predicting cellular responses during drug development, to informing clinical trial design, to building adaptable AI models that generalize across tissues and perturbations.

These kinds of datasets and collaborative events help bridge the gap between raw data and real-world biological insight. They also highlight how foundational models, when applied thoughtfully, can accelerate progress across the entire drug discovery pipeline.

Thank You

A huge thanks to the team at Tahoe Therapeutics for organizing the hackathon and for making such valuable data accessible to the community. Events like this remind us of the power of open science and the breakthroughs that small, interdisciplinary teams can achieve in a short time.