Foundation Models for Biology Seminar Series (FM4Bio)

The Foundation Models for Biology (FM4Bio) Seminar Series by GenBio AI explores the applications of multiscale foundation models in solving complex biological challenges. These models integrate multimodal biological data—DNA, RNA, proteins, and single-cell information—to predict, simulate, and program biology at different levels.

As part of GenBio AI’s broader vision to develop the world’s first AI-Driven Digital Organism (AIDO), these seminars highlight the latest advancements in using foundation models to uncover biological insights, accelerate drug discovery, and personalized medicine.

Our seminars bring together experts working at the intersection of AI and biology to present groundbreaking work and explore how foundation models can bridge the gap between molecular data and systems biology.

The series is hosted by Caleb Ellington and offers an opportunity for the research community to stay informed and engaged with cutting-edge innovations.

👉 Watch the full seminar playlist on YouTube: YouTube FM4Bio Seminar Series.

Register For Our Next Session

---

Schedule

All upcoming and past talks in our series.

🚀 Upcoming Talks

TBA

🧬 Past Talks

✅Lightweight, Interpretable Supervised and Self-Supervised Deep Learning Models of Regulatory DNA (Blog and Video in progress)
Presenter: Anshul Kundaje
Date: March 24- 9:00 AM PT
Summary: Anshul Kundaje will present three complementary frameworks from Stanford, ChromBPNet, DART-Eval, and ARSENAL, demonstrating that lightweight, interpretable models consistently outperform large DNA foundation models at decoding regulatory DNA, predicting variant effects, and designing regulatory sequences, all while requiring far less compute.

✅ Large Models for Single-Cell Omics and Drug Discovery: Data, Pretraining, and Closed-Loop Environment
Presenter: Haotian Cui
Date: September 17 – 9:00 AM PT
Summary: Haotian Cui will present scGPT, a generative transformer trained on over 33 million single-cell profiles, and LUMI-lab, a closed-loop platform for model-guided mRNA delivery. Together, these advances enable scalable virtual cell modeling, multimodal integration, and active learning for therapeutic discovery.

✅ Predicting Cellular Responses to Perturbation Across Diverse Contexts With STATE
Presenter: Abhinav Adduri
Date: August 5, 2025 – 9:00 AM PT
Summary: Abhi will present STATE, a machine learning architecture that predicts cellular responses to perturbations across diverse contexts with over 50% improved accuracy, enabling scalable development of virtual cell models.

✅ Egret-1: Pretrained Neural Network Potentials for Efficient and Accurate Bioorganic Simulation
Presenter: Elias L. Mann
Date: June 18, 2025
Summary: Eli presented Egret-1, a family of large pretrained neural network potentials designed to enable fast and accurate quantum-level simulations for bioorganic and biomolecular systems.

✅ Biomni: A General-Purpose Biomedical AI Agent
Presenter: Kexin Huang
Date: June 4, 2025
Summary: Kexin presented Biomni, a general-purpose biomedical AI agent designed to automate complex research tasks across biomedicine using LLMs, code execution, and real-world data.

✅ scPRINT: pre-training on 50 million cells allows robust gene network predictions
Presenter: Jérémie Kalfon
Date: May 16, 2025
Summary: Jérémie showcased scPRINT, a large cell foundation model pre-trained on 50 million cells to advance robust gene network inference and interpretability in single-cell biology.

✅ Mapping Cell Fate Transition in Space and Time
Presenter: Joshua D. Welch
Date: April 14, 2025
Summary: Joshua showcased TopoVelo, a probabilistic model that infers spatial and temporal dynamics of cell fate transitions by extending RNA velocity to spatial transcriptomic data.

✅ Mapping Cells Through Time and Space With Moscot
Presenter: Dominik Klein
Date: March 26, 2025
Summary: Dominik showcased moscot, an optimal transport framework that maps cellular dynamics across space and time, integrating multimodal single-cell data to reveal new insights into pancreas biology.

✅ How Molecules Impact Cells: Unlocking Contrastive PhenoMolecular Retrieval
Presenter: Philip Fradkin
Date: March 3, 2025
Summary: Phil introduced MolPhenix, a contrastive learning model that improves the retrieval of active molecules by aligning phenomic (cell-painting) data with molecular structures.