Alphamod

Alphamod integrates AlphaFold2 and MODELLER to refine protein tertiary structure predictions and improve model accuracy for evolutionary analysis, protein–protein interaction studies, and rational drug design.

Key Features:

• Integration of AlphaFold2 and MODELLER: Alphamod generates initial models with AlphaFold2's deep learning predictions and refines them using MODELLER's template-based modeling to incorporate known structural templates.
• Comprehensive Quality Assessment: Alphamod applies multiple quality assessment tools and synthesizes their outputs into a composite score, BORDASCORE, which correlates with the Global Distance Test Total Score (GDT_TS) to aid model selection when no reference structure is available.
• Validation and Performance: Alphamod was validated on two datasets comprising 72 targets previously used to assess AlphaFold2, using averaged GDT_TS across generated structures and pairwise comparisons, and reported approximately a 34% increase in accuracy in unsupervised setups and outperforming AlphaFold2 in 18% of supervised cases.
• Flexibility for Future Enhancements: The pipeline is designed to allow integration of additional data sources and AI-based algorithms to further improve prediction reliability.

Scientific Applications:

• Evolutionary Biology: More accurate tertiary structures support analysis of evolutionary relationships within protein families.
• Protein-Protein Interaction Studies: Improved structural models facilitate examination of interaction interfaces and complex formation.
• Rational Drug Design: Accurate protein models aid identification of drug targets and structure-based ligand design.

Methodology:

Initial prediction with AlphaFold2 produces three-dimensional conformations from amino acid sequences; refinement with MODELLER applies template-based modeling to improve those predictions; quality assessment tools are aggregated into BORDASCORE and evaluated against GDT_TS.