dkNET Webinar "Towards a Virtual Cell - the Need for Actionable, Robust Perturbation Models"

Join dkNET Webinar on Thursday, March 12, 2026, 11 am - 12 pm PT

Presenter: Prof. Fabian Theis, PhD, Institute of Computational Biology, Helmholtz Center Munich, Germany; TUM Schools of Computation, Information and Technology & Life Sciences, Technical University of Munich, Germany

Abstract

The vision of a virtual cell, i.e. a computational model that predicts cellular responses to genetic or pharmacological interventions, has been around for decades. Recent large-scale foundation models in single-cell genomics promise to accelerate this vision, yet they often remain descriptive and fall short of robust, intervention-ready prediction.

In this talk, I will highlight our contributions toward closing this gap: from atlas-scale representation learning and multimodal integration to foundation models such as NicheFormer, which learn transferable embeddings across tissues and contexts. Building on this, I will present CellFlow, our generative perturbation framework that models how drugs, gene edits, or differentiation protocols shift cellular states and enables in silico experimental design.

I will argue that achieving virtual cells requires not only scale, but carefully designed perturbation data and robust generative models that move from observing biology to shaping it.

Dial-in Information: 
https://uchealth.zoom.us/webinar/register/WN_ah0G3w3YQwqoV15WQYKe7g

Date/Time: Thursday, March 12, 2026, 11 am - 12 pm PT

Upcoming webinars schedule: https://dknet.org/about/webinar