MULTICOM2

MULTICOM2 predicts protein tertiary structures by integrating template-based and template-free modeling to determine protein folds and inter-residue distances for structural biology and related applications.

Key Features:

• Integration of TBM and FM Methods: Combines template-based modeling (TBM) and template-free modeling (FM), using sequence alignment tools with deep multiple sequence alignments to identify structural templates and improve performance over MULTICOM1.
• Template-Free Modeling via Deep Learning: Performs template-free (ab initio) modeling for domains without templates by predicting inter-residue distances with the DeepDist algorithm to reconstruct tertiary structures.
• Performance in CASP14 Experiment: Achieved average TM-scores of 0.720 for 58 TBM domains and 0.514 for 38 FM and FM/TBM domains in CASP14, correctly predicting the fold for 76 domains with reported single-prediction success rates of 95% for regular domains and 55% for hard domains, and reported success rate changes to 3% for both regular and hard domains when five predictions were made per domain.
• High Accuracy Across Modeling Methods: Found pure template-free modeling accuracy to be very close to combined TBM and FM results, indicating deep learning distance-based template-free modeling can effectively replace template-based approaches for many targets.
• Top Performer in CASP14: Server predictors MULTICOM-HYBRID, MULTICOM-DEEP, and MULTICOM-DIST ranked among the top 20 automated server predictors in CASP14, and combining group predictors placed MULTICOM-HYBRID at rank no. 5.

Scientific Applications:

• Structural Biology: Provides predicted tertiary structures for structural biology analyses of known and novel proteins.
• Drug Discovery: Generates tertiary-structure models useful for structure-based drug discovery.
• Protein Function Interpretation: Aids understanding of protein function at the molecular level through predicted structures.

Methodology:

Uses sequence alignment with deep multiple sequence alignments to identify templates for TBM, applies the DeepDist deep-learning algorithm to predict inter-residue distances for template-free modeling, and reconstructs tertiary structures from predicted distances while integrating TBM and FM results.