DeepUMQA-X

DeepUMQA-X combines single-model protocols for various evaluation metrics with a consensus strategy to estimate protein model accuracy for single-chain and complex structures.

Key Features:

• Single-model protocols: Outperformed all other single-model methods in accuracy assessment in the CASP16 EMA blind test.
• Consensus strategy: Integrates with single-model approaches to bridge the performance gap between consensus methods and single-model methods.
• Multi-input support: Accepts multiple protein single-chain or complex models as input simultaneously.
• Comprehensive accuracy estimates: Provides overall, interface, and residue accuracy estimates for each submitted model.
• Blind test validation: Achieved top performance across nearly all CASP16 EMA tracks including QMODE1, QMODE2, QMODE3, and self-assessment, and ranked first in a one-year CAMEO-QE blind test from 9 June 2023 to 1 June 2024.

Scientific Applications:

• Protein structure prediction refinement: Enables accurate scoring, ranking, and selection of protein models from large-scale sampling in the post-AlphaFold2 era.
• Complex structure quality assessment: Evaluates accuracy of predicted protein-protein complex interfaces and overall complex models.
• Model accuracy estimation: Supports estimation of model accuracy for both single-chain and multimeric protein structures in blind prediction benchmarks.

Methodology:

DeepUMQA-X utilizes single-model methods combined with a consensus strategy for protein model accuracy estimation.