Engineering selective amyloid precursor protein inhibitors by machine learning and deep mutational scanning

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Deep mutational scanning (DMS) maps protein-protein interactions (PPIs) but has limitations in covering multi-mutant variants.
• Amyloid precursor protein inhibitors (APPIs) target serine proteases like mesotrypsin and kallikrein-6 (KLK6), which are implicated in human diseases.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting APPI binding selectivity towards mesotrypsin and KLK6.
• To identify novel APPI variants with enhanced binding selectivity and therapeutic potential.

Main Methods:

• Training ML models on existing DMS data for APPI interactions with mesotrypsin and KLK6.
• Combining ML models to predict the binding selectivity of single and double-mutant APPI variants.
• Experimental validation using yeast-surface display and enzyme inhibition assays.

Main Results:

• Achieved a high Pearson correlation (0.937) between predicted and experimentally determined log2 selectivity enrichment ratios.
• Identified epistatic interactions influencing protease selectivity.
• Discovered highly selective APPI variants, including the most potent mesotrypsin inhibitor to date.

Conclusions:

• DMS combined with ML provides a robust framework for predicting PPI selectivity.
• This integrated approach accelerates the discovery and optimization of therapeutic protein variants.
• The identified selective APPI variants hold promise for treating diseases associated with mesotrypsin and KLK6.