Degradation graphs reveal hidden proteolytic activity in peptidomes

Area of Science:

• Proteomics and biochemistry
• Computational biology and bioinformatics

Background:

• Protein degradation is a crucial biological process that modifies the proteome and produces bioactive peptides.
• Current peptidomics and degradomics analyses often overlook the sequential nature of peptide generation, treating peptides as isolated products.

Purpose of the Study:

• To introduce a novel probabilistic framework, degradation graphs, for modeling proteolysis as a network of sequential cleavage events.
• To improve the accuracy of quantifying proteolytic activity and identify protease-specific signatures.

Main Methods:

• Developed degradation graphs, a directed acyclic network model representing proteolysis with explicit absorption.
• Inferred graph weights using gradient descent or linear-flow optimization from single-snapshot peptidome data.
• Quantified proteolytic flows and corrected biases in conventional quantification.

Main Results:

• Failure to model downstream peptide trimming leads to 3-4 fold underestimation of upstream proteolytic activity.
• Degradation graphs provide graph-structured features for machine learning models.
• Machine learning models utilizing these features can capture protease-specific signatures based on graph topology and sequence context.

Conclusions:

• Explicit modeling of protein degradation pathways via degradation graphs offers a practical and mechanistic approach to peptidomics.
• This framework bridges degradomics and peptidomics, enabling more interpretable and accurate analysis of protein breakdown.
• The method enhances the ability to identify protease activity and understand complex proteolytic events.