Lessons from the Virtual Cell Challenge: When Simple Models Win

In this session, we'll share our experience from the Arc Institute Virtual Cell Challenge, where the Turbine Mean Predictors team placed 15th of 337 participating teams overall and 3rd on the newly introduced generalist leaderboard.

We'll walk through why a simple linear, biology-informed model trained on pseudo-bulk data was competitive with far more complex deep learning and foundation-model approaches.

Beyond results, we'll discuss what the challenge revealed about single-cell vs. pseudo-bulk modeling, feature design, and the pitfalls of metric-driven optimization. Finally, we'll reflect on what these lessons mean for building robust, scalable virtual cell models at Turbine - and for future benchmarking challenges.

=== ENTRY DETAILS ===

- QR code with entry information will be available soon, in the "Photos" section of this event page.
- Gate closes at 18:15 - no late entries.