Essential gene prediction using limited gene essentiality information-An integrative semi-supervised machine learning...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Essential gene prediction identifies genes critical for organism survival.
• Machine learning (ML) aids gene essentiality prediction, but current methods struggle with limited experimental data.
• Less-explored, disease-causing organisms often lack sufficient data for accurate essential gene annotation.

Purpose of the Study:

• To develop a novel ML pipeline for predicting essential genes in organisms with limited experimental data.
• To improve the annotation of essential genes for understudied, pathogenic organisms.
• To provide a robust strategy for identifying potential therapeutic targets.

Main Methods:

• A pipeline combining unsupervised feature selection, Kamada-Kawai dimension reduction, and Laplacian Support Vector Machine (LapSVM) was developed.
• A novel Semi-Supervised Model Selection Score, analogous to auROC, was proposed for model selection with limited data.
• Genome-scale metabolic networks were utilized for predicting essential and non-essential genes.

Main Results:

• The pipeline achieved high accuracy (auROC > 0.85) in essential gene prediction, even with as little as 1% labeled data.
• Unsupervised feature selection and dimension reduction revealed distinct clustering patterns for essential and non-essential genes.
• The method demonstrated high accuracy and universality across Eukaryotes and Prokaryotes, including Leishmania sp.

Conclusions:

• The proposed ML pipeline effectively predicts essential genes using limited labeled data, addressing a critical gap in bioinformatics.
• This approach facilitates essential gene annotation for poorly characterized organisms, aiding in the identification of novel therapeutic targets.
• The strategy offers a valuable tool for antibiotic and vaccine development against parasitic diseases.