ColabFold

ColabFold accelerates prediction of protein structures and complexes by integrating MMseqs2 rapid homology searches with AlphaFold2 and RoseTTAFold structure prediction models.

Key Features:

• MMseqs2 integration: Integrates MMseqs2 for rapid homology searches to accelerate sequence search steps.
• Structure prediction models: Uses AlphaFold2 and RoseTTAFold for protein structure and complex prediction.
• Speed and throughput: Delivers homology searches reported to be 40–60× faster than traditional methods, enabling prediction throughput of approximately 1,000 protein structures per day on a single GPU server.
• Expandable profile databases and MSA generation: Employs MMseqs2 expandable profile databases to generate diverse multiple sequence alignments essential for accurate structure prediction.
• Environmental databases: Incorporates novel environmental databases to expand sequence space available for alignments and predictions.
• Optimized model utilization: Optimizes model usage to maximize prediction throughput and computational resource efficiency.

Scientific Applications:

• Protein structure prediction: Prediction of tertiary structures for individual proteins using AlphaFold2 or RoseTTAFold informed by MSAs.
• Protein complex modeling: Prediction of protein complexes leveraging the integrated structure prediction models.
• High-throughput structural studies: Large-scale structural prediction campaigns enabled by accelerated homology searches and optimized compute usage.
• Sequence-space expansion: Use of environmental databases to include remote homologs and environmental sequences in alignment and prediction workflows.

Methodology:

Performs MMseqs2 rapid homology searches using expandable profile databases to generate diverse multiple sequence alignments, then applies AlphaFold2 or RoseTTAFold with optimized model utilization for structure and complex prediction.