bbcontacts: prediction of β-strand pairing from direct coupling patterns.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Reliable de novo protein models require large multiple sequence alignments (MSAs) of homologous sequences.
• Statistical network analysis of MSAs reveals residue-residue contacts from correlations between sequence columns.
• Exploiting structural regularity in beta-strands can improve contact prediction and reduce MSA size requirements.

Purpose of the Study:

• To develop a method for reliably predicting beta-strand to beta-strand (β-β) contacts in proteins.
• To reduce the dependency on large MSAs for accurate protein structure modeling.
• To improve the precision and recall of contact prediction compared to existing methods.

Main Methods:

• Developed bbcontacts software utilizing two hidden Markov models (HMMs) to detect characteristic patterns of β-β contacts.
• Modeled β-bulges as indel states within the HMM framework.
• Employed predicted secondary structure rather than true secondary structure for contact prediction.

Main Results:

• bbcontacts achieved 50% precision at 50% recall for β-β contacts using predicted secondary structure.
• Using true secondary structure, bbcontacts reached 64% precision at 64% recall.
• These results surpass existing tools, which achieve approximately 45% precision at 45% recall.

Conclusions:

• bbcontacts enhances the prediction of β-β contacts, a crucial step in de novo protein structure modeling.
• The method demonstrates improved performance, particularly for proteins with smaller MSAs.
• This advancement contributes to more accurate and efficient protein structure prediction.