Data-driven supervised learning of a viral protease specificity landscape from deep sequencing and molecular...

Area of Science:

• Biophysics
• Molecular Biology
• Computational Biology

Background:

• Protein-peptide interactions are crucial for molecular recognition specificity.
• Understanding how structure and energetics at interfaces shape these landscapes is challenging.

Purpose of the Study:

• To comprehensively map the specificity landscape of the hepatitis C virus (HCV) NS3/4A protease.
• To establish a sequence-energetics-function relationship for protease substrate recognition.

Main Methods:

• Yeast-based library screening and deep sequencing to measure substrate cleavability.
• Structure-based modeling and supervised machine learning (support vector machine) to predict substrate variants.
• Graph-theoretic analyses to study sequence space clustering.

Main Results:

• Generated a predictive model for 3.2 million substrate variants of the HCV protease.
• Identified novel HCV protease substrates and revealed clustering of cleavable/uncleavable motifs.
• Observed similar clustering for drug-resistant variants, suggesting a role for energetics in functional landscapes.

Conclusions:

• The developed approach enables comprehensive mapping and redesign of protease specificity landscapes.
• This method is applicable to a wide range of proteases, including human enzymes.
• Residue-level energetics may play a role in shaping functional landscapes, aligning with viral quasispecies theory.