TL;DR
In this talk, Joshua D. Welch presents his paper “Mapping Cell Fate Transition in Space and Time” at the Foundation Models for Biology Seminar Series by GenBio AI. He introduces TopoVelo, a model that uses spatial transcriptomics and RNA velocity to track how cells change and move within tissues. TopoVelo helps scientists better understand tissue development, cell migration, and the hidden dynamics of early life by combining gene expression data with physical location.
Introduction
In biology, timing and context are everything. While single-cell RNA sequencing gives us powerful snapshots of what a cell is doing, it strips away when and where that cell exists. Yet tissues are dynamic: cells migrate, differentiate, and interact in space and time. How can we reconstruct those transitions and understand the full picture?
In a recent Foundation Models for Biology Seminar, Professor Josh Welch shared a compelling approach to this problem. His team’s model, TopoVelo, brings spatial transcriptomics and RNA velocity together to model how cells change across both physical space and developmental time. Rather than just looking at gene expression in isolation, TopoVelo enables researchers to map cell transitions in four dimensions: space, time, gene expression, and interaction.
Full Talk
Watch the full talk below:
The Problem: Capturing Complexity in a Moving System
Biologists often want to understand how cells transition from one state to another, how stem cells become neurons, or how immune cells migrate into tissues. But most existing tools treat cells as isolated points in gene space. They lack spatial resolution, ignore neighboring influence, and typically assume static or simple dynamics.
What Welch’s team realized is that cell state isn’t just about the cell itself; it’s also shaped by its position in a tissue and by the other cells around it. With new technologies in spatial transcriptomics, it became possible to build a model that accounts for all three: where cells are, what they express, and who they interact with.
What Is TopoVelo?
TopoVelo stands for Topological Velocity, and it’s designed to extend RNA velocity into space by treating tissues as spatially coupled systems. Built on a graph-based variational autoencoder, the model uses spliced and unspliced RNA, which are signals of gene activity at different stages, to estimate the future state of each cell.
But unlike traditional velocity methods, it models this change while factoring in spatial proximity and neighbor influence.
The result is a dynamic tissue map, where arrows represent predicted trajectories of cell movement or differentiation over time. The method scales to large tissue datasets, incorporates spatial relationships through graph attention networks, and provides interpretable outputs like velocity fields and cell influence scores.
Applications: From Mouse Embryos to Human Organoids
Welch’s team applied TopoVelo across diverse biological systems, each demonstrating how tissues grow, differentiate, or reorganize over time.
- In developing mouse brains, TopoVelo traced the migration of neurons outward from progenitor zones, matching known neurodevelopmental patterns.
- In stereotactic embryonic data, it captured radial expansion in the gut and lungs.
- In the thymus, it revealed inward migration patterns of differentiating thymocytes.
- In a 3D unrolled neural tube model, it identified the exact locations of known neural tube closures using only gene expression and spatial data.
Perhaps most striking was its application to human embryoid bodies. These lab-grown organoids simulate early development. TopoVelo showed that cells likely differentiate outward from central lumens and identified areas with high influence, possibly linked to WNT signaling, a known developmental pathway.
Going Beyond Description
What makes TopoVelo more than just a visualization tool is its generative modeling power. It doesn’t just describe what is observed. It builds simulations of what could happen under different biological conditions. This opens doors to:
- Simulating how cells might behave under hypothetical changes
- Predicting differentiation direction and speed
- Identifying influential regions or cells that shape their neighborhood
In early tests, TopoVelo’s velocity estimates even matched known migration rates from in vivo imaging, offering validation of its biological accuracy.
Final Thoughts
TopoVelo marks a significant advancement in our ability to model living tissues. By combining spatial transcriptomics, RNA velocity, and graph-based learning, it brings us closer to reconstructing biology not just in static snapshots but in continuous, interpretable motion.
For further details, check out the full paper on TopoVelo.
Join us in our mission to push the frontiers of AI-driven biology and strive to make a lasting impact on medicine, biotechnology, and human health. We are hiring across teams. Visit our Careers page to learn more and apply. Follow us on X, YouTube, and LinkedIn.
