SADA

SADA assembles full-chain multi-domain protein structures by identifying structural analogues from a multi-domain protein structure database and integrating deep learning-predicted inter-residue distance potentials to improve modeling accuracy.

Key Features:

• Structural Analogue-Based Approach: Constructs and searches a database of multi-domain protein structures to identify structural analogues and detect full-chain analogues from individual domain models.
• Deep Learning Integration: Predicts inter-residue distance potentials with deep learning and incorporates them into an energy function to guide domain assembly.
• Two-Stage Differential Evolution Algorithm: Uses a two-stage differential evolution optimization to simulate and refine domain assemblies for accurate full-chain modeling.

Scientific Applications:

• Multi-domain protein modeling: Improves modeling of inter-domain interactions and full-chain structures for multi-domain proteins.
• Benchmark performance: In a benchmark of 356 proteins, achieved average TM-score improvements of 8.1% over DEMO and 27.0% over AIDA.
• Human protein assembly: On assembly of 293 human multi-domain proteins, produced models with a 1.1% higher TM-score compared to AlphaFold2.

Methodology:

Initial model construction from structural analogues detected in a multi-domain protein structure database, followed by domain assembly simulation using a two-stage differential evolution algorithm guided by an energy function informed by deep learning-predicted inter-residue distance potentials.