Google DeepMind Unveils AlphaGenome AI Model That Predicts DNA Function at Single Base-Pair Resolution

Google DeepMind releases AlphaGenome, a groundbreaking AI model that predicts DNA function and regulatory variant effects at single base-pair resolution by analyzing up to 1 million base pairs, with research code and pre-trained weights now available for non-commercial use.

• Google DeepMind releases AlphaGenome, a unified DNA sequence model that analyzes up to 1 million base pairs to predict regulatory variant effects and genome function at single base-pair resolution
• The research code repository provides JAX implementation, API with variant scorers, and dataset loaders, with pre-trained model weights available through Kaggle and Hugging Face under non-commercial terms
• AlphaGenome requires high-end hardware including NVIDIA H100 GPUs for inference and TPU v3+ for training, while delivering predictions across gene expression, splicing patterns, chromatin features, and contact maps