Stanford Releases ENCODE GRAMMAR to Decode Genomic Regulatory Logic via Deep Learning
anshulkundaje · x · 2026-08-05
Stanford's Kundaje Lab released ENCODE GRAMMAR, a deep learning model resource designed to decode the DNA sequence logic of regulatory elements in the human genome.
- Model Scale: Contains 3,865 experiment-specific model sets, including 2,339 BPNet TF ChIP-seq sets, 1,512 ChromBPNet chromatin accessibility sets, along with models for transcription initiation and reporter assays.
- Architecture: Uses a lightweight, local BPNet architecture that predicts genome-wide biochemical profiles at single-basepair resolution using up to 2kb of local DNA sequence context. Each experiment is represented by an ensemble of models trained on different chromosome splits to ensure robust predictions and estimate uncertainty.
- Interpretation Toolkit: Provides tools like DeepLIFT/DeepSHAP (to score base-level contributions) and TF-MoDISco (to discover predictive motifs), bridging the gap between model predictions and the underlying driving sequence patterns.
- Case Study: The accompanying blog demonstrates how to integrate annotations from multiple GRAMMAR models to decode an accessible regulatory element of the MYC oncogene in a leukemia cell line.
More from Research
- Harvard, MIT, Stanford Launch MatrAIx: 8.3B Persona Agents for AI Evaluation — BrihiJ · 2026-08-06
- Dongfang Starlink Launches AI Hyperspectral Satellites, Plans 2M Lunar Robots for Data Center — 新智元 · 2026-08-06
- USTC Study: AI Agents Achieve Only 3.3% Zero-Intervention Execution in Robot Labs — 新智元 · 2026-08-06
- 23-Year-Old OpenAI Alignment Researcher Joins Brain-Computer Startup Conduit — 新智元 · 2026-08-06
- New OGM Neural Network Architecture for Disease Risk Prediction — anshulkundaje · 2026-08-06
- CoCoEvolve: A New Framework for Cross-Representation Consistency in Charts, Tables, and Code — Xuehang Guo · 2026-08-06