Making Use of Averaging Methods in MODELLER for Protein Structure Prediction

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Modeling

Background:

• Advanced protein structure prediction methods like AlphaFold 2 excel at static structures but struggle with dynamic features.
• Homology-based modeling offers a cost-effective alternative for protein structure prediction.
• The MODELLER algorithm generates intermediate models during simulated annealing, exploring conformational states.

Purpose of the Study:

• To investigate if averaging intermediate MODELLER models improves structural quality.
• To assess the efficacy of statistical potential ranking for selecting optimal models.
• To explore the impact of single vs. multiple templates on the averaging approach.
• To determine if MODELLER's ensemble phase can capture low-energy holo conformations.

Main Methods:

• Examined 225 target-template pairs using MODELLER (version 10.5).
• Extracted and averaged representative intermediate models based on statistical potentials.
• Evaluated model quality improvements and ranking reliability.

Main Results:

• Averaging intermediate models yielded minor but notable improvements in structural similarity to native structures.
• The averaging procedure was particularly beneficial in low sequence identity regions and for holo protein conformation generation.
• A significant fraction of averaged configurations showed higher similarity than MODELLER's final output.

Conclusions:

• Averaging intermediate MODELLER models can enhance structural model quality, especially in challenging homology modeling scenarios.
• Statistical potential ranking aids in selecting improved models.
• The method shows promise for capturing holo protein conformations, an area where traditional methods often fall short.