PON-All: Amino Acid Substitution Tolerance Predictor for All Organisms

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Interpreting genetic variations is crucial for clinical diagnosis but challenging.
• Existing variant prediction tools have limited performance and species applicability.
• A need exists for a versatile predictor applicable across diverse organisms.

Purpose of the Study:

• To develop a novel machine learning tool for predicting amino acid substitution pathogenicity in any organism.
• To create a predictor that overcomes limitations of existing human-centric and species-specific tools.
• To provide predictions for pathogenic, benign, and unknown significance variants.

Main Methods:

• Developed PON-All, a machine learning model trained on human, animal, and plant genetic variants.
• Implemented two versions: one utilizing Gene Ontology (GO) annotations and another without.
• Evaluated performance using blind tests with accuracy and Matthews Correlation Coefficient (MCC) metrics.

Main Results:

• PON-All achieved high accuracy (0.913) and MCC (0.827) with GO annotations.
• Without GO features, accuracy was 0.856 and MCC was 0.712.
• Demonstrated superior performance compared to other variant prediction tools, particularly for human and plant variants.

Conclusions:

• PON-All offers a powerful and broadly applicable solution for variant pathogenicity prediction.
• The tool's performance is robust, even without species-specific GO annotations.
• PON-All represents a significant advancement in interpreting genetic variations across diverse species.