Stanford's Anshul Kundaje fires 9-tweet rebuttal at study claiming TF binding lacks selectivity
anshulkundaje · x · 2026-11-19
Stanford genomics researcher Anshul Kundaje posted a 9-part thread criticizing a new study that used his openly released BPNet-style models and data resources to argue transcription factors do not selectively bind genomic targets. While welcoming the fast open release, he disputes the conclusions: the analysis (1) does not test selectivity—why some motifs are bound and others are not; (2) uses short-context models like BPNet trained on peaks with GC-matched backgrounds, which are ill-suited for that question—genome-wide long-context models would perform better; (3) is overly conservative for testing whether TF A binding depends on other TFs using directional assays like ChIP-seq. He cites extensive prior evidence of complex TF dependence (direct interactions, composites, hard/soft syntax often chromatin-mediated, super-additive/multiplicative effects) and notes other localization signals like IDRs and non-DBDs, which his models capture as predictive non-canonical features. His team is running a comprehensive re-analysis over the coming months and has reached 'a very different conclusion,' with a detailed preview promised in about a week.
Related event: Stanford's Kundaje fires nine-tweet rebuttal over TF cooperativity study(10 posts)→
More from Research
- Shanghai AI Lab's GameRealm: 120M frames of gameplay data for visual foundation models — Shanghai-AI-Laboratory · 2026-11-19
- KradleBench launches as a dynamic multi-agent benchmark pitting LLMs, world models and humans against each other — lmoroney · 2026-11-19
- KradleBench Launches: Minecraft Legend SethBling Dethroned by Model Astra 6.2 — fchollet · 2026-11-19
- Transformer Lab unveils Primus, a 'society of researchers' of AI agents, and is hiring — carsonfarmer · 2026-11-19
- New LessWrong Analysis: LLMs Seek the Grader's Score, Not the Actual Goal — 1a3orn · 2026-11-19
- tabgen-eval: an open-source eval framework and benchmark for synthetic tabular data — LucaAmb · 2026-11-19