How to Accelerate Protein Structure Prediction at Proteome-Scale

NVIDIA, Google DeepMind, EMBL-EBI, and Seoul National University collaborated to extend the AlphaFold Protein Structure Database (AFDB) beyond monomeric structures to proteome-scale quaternary structures, predicting over 31 million homo- and heteromeric protein complexes across 4,777 proteomes, with 1.8 million high-confidence homodimer structures now publicly available. This was achieved by leveraging AlphaFold-Multimer accelerated via NVIDIA H100 GPUs, MMseqs2-GPU for MSA generation, NVIDIA TensorRT, and NVIDIA cuEquivariance, implemented through a modular, decoupled pipeline separating MSA generation and structure inference with optimized SLURM orchestration to maximize GPU utilization. The result enables proteome-scale, interaction-aware structure prediction by integrating STRING-derived interaction data and robust quality calibration, making high-confidence protein complex structures accessible for downstream applications in systems biology, drug discovery, and generative protein modeling.

Related

• Running Large-Scale GPU Workloads on Kubernetes with Slurm
• NestPipe: Large-Scale Recommendation Training on 1,500+ Accelerators via Nested Pipelining
• Nvidia published DWDP (Distributed Weight-Data Parallelism), a new inference parallelism strategy focused on prefill. It sounds slightly ins…
• New course: Efficient Inference with SGLang: Text and Image Generation, built in partnership with LMSys @lmsysorg and RadixArk @radixark, an…
• CodecSight: Leveraging Video Codec Signals for Efficient Streaming VLM Inference

Source: hardware