SVM-based prediction of the calpain degradome using Bayes Feature Extraction

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• Calpains are calcium-dependent cysteine proteases linked to various diseases, including cancer and neurodegeneration.
• The full extent of proteins targeted by calpains (the calpain degradome) remains incompletely understood despite significant research.

Purpose of the Study:

• To identify novel sequence features predictive of calpain cleavage sites.
• To develop and validate computational models for predicting calpain cleavage sites.
• To investigate the role of calpains in regulating receptor tyrosine kinase (RTK) activity.

Main Methods:

• Analysis of a dataset comprising 341 experimentally verified calpain cleavage sites.
• Sequence analysis to identify residue propensities flanking cleavage sites.
• Development of machine learning models, including Support Vector Machines and Bayes Feature Extraction, for cleavage site prediction.
• Validation of models using independent test sets.
• Prediction of calpain cleavage sites on receptor tyrosine kinase proteins.

Main Results:

• Discovery of novel residue propensities in regions surrounding calpain cleavage sites.
• Machine learning models achieved high predictive performance, with Area Under the Receiver Operating Characteristic Curve (AROC) scores from 0.79 to 0.93 and accuracy from 71% to 86%.
• Identification of potential calpain cleavage sites within critical regulatory domains of receptor tyrosine kinases.

Conclusions:

• The developed computational models offer a robust method for predicting calpain cleavage sites.
• Calpains may play a direct role in regulating receptor tyrosine kinase activity.
• This regulation by calpains could impact cellular pathways governing cell survival and cell death.