Kundaje: perturb-seq prediction models are not automatically virtual cells
anshulkundaje · x · 2026-08-29
Computational biologist Anshul Kundaje raised a pointed critique amid discussion of the new CD4+ T cell perturb-seq dataset:
- Such data is of course very useful, as the paper itself demonstrated;
- But the premise that training and testing models on perturb-seq data will deliver virtual cells is fundamentally flawed — models predicting perturbation effects are not automatically virtual cells.
A substantive methodological challenge to the current AIVC roadmap.
More from AGI Musings
- Game studios can sell privately hosted realistic games to AI labs as training environments — dejavucoder · 2026-10-15
- Retellings Use New Perspectives to Replace Romance with Therapeutic Diagnosis — eigenrobot · 2026-10-15
- CHI Reviewer: AI Slop Is Real, but Unusually Rigorous Papers Still Getting Through — IanArawjo · 2026-10-15
- ASML Sells Zero Lithography Machines into Europe as US Hosts 75%+ of AI Compute — NinaDSchick · 2026-10-15
- Into the Unknown: CMU benchmark finds 114 of 135 AI-written research papers fail integrity checks — arankomatsuzaki · 2026-10-15
- LLMs-as-Media Erode Producer Power; Author Ties It to Securities Reform — curious_vii · 2026-10-15