Homology Modeling in the Twilight Zone: Improved Accuracy by Sequence Space Analysis.

Area of Science:

• Structural bioinformatics
• Protein evolution
• Computational biology

Background:

• Protein sequence divergence can challenge structural conservation and homology modeling.
• The 'twilight zone' of sequence divergence requires specialized methods for reliable protein modeling.
• Identifying conserved structural templates is crucial for homology modeling, especially for distantly related proteins.

Purpose of the Study:

• To establish a method for reliable homology modeling of proteins with significant sequence divergence.
• To investigate the relationship between sequence and structure conservation in protein evolution.
• To demonstrate improved protein modeling through expert-curated data and sequence mining.

Main Methods:

• Analysis of sequence-structure similarity relationships in protein families.
• Application of conventional threading methods and deep learning approaches (e.g., AlphaFold).
• Sequence database mining combined with Multidimensional Scaling (MDS) to identify intermediary sequences.
• Phylogenetic analysis to infer homology and structural conservation.
• Comparison of automated web server models with expert-customized models.

Main Results:

• A limit for confident structural conservation and reliable homology modeling based on sequence divergence was identified.
• Intermediary sequences were successfully identified between the plethodontid receptivity factor isoform 1 (PRF1) and its structural template, CNTF.
• Phylogenetic and MDS analyses provided crucial links, enabling reliable homology modeling for PRF1.
• Expert-guided modeling improved upon automated models generated by web servers.

Conclusions:

• Sequence mining and MDS are effective strategies to overcome the challenges of the 'twilight zone' in homology modeling.
• Identifying intermediary sequences is key to inferring homology and structural conservation between distantly related proteins.
• Expert input significantly enhances the accuracy and reliability of protein structure modeling, particularly for novel or weakly related protein families.