Evaluation of machine learning classifiers for predicting essential genes in Mycobacterium tuberculosis strains

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Essential genes are vital for bacterial survival and are key targets for antimicrobial drug development.
• Understanding the biological mechanisms of bacterial cells relies on accurate identification of essential genes.
• Mycobacterium tuberculosis poses a significant global health challenge, necessitating novel therapeutic strategies.

Purpose of the Study:

• To accurately predict essential genes in Mycobacterium tuberculosis.
• To identify key genomic features predictive of essentiality.
• To develop and evaluate a novel deep neural network (DNN) model for essential gene prediction.

Main Methods:

• A genome extraction program was utilized to map and match essential gene information from the ePath and NCBI databases for 20 strains of Mycobacterium tuberculosis.
• A Genetic Algorithm was employed to select a subset of key features from 14 genome sequence-based features.
• The selected key features were used to train, validate, and test three classifiers: DNN, Decision Tree (DT), and Support Vector Machine (SVM), with an 80%, 10%, and 10% data split, respectively.

Main Results:

• The proposed DNN model achieved a high Area Under the Curve (AUC) of 0.98, significantly outperforming DT (AUC=0.88) and SVM (AUC=0.82).
• The use of a key feature subset enhanced classifier generalizability and demonstrated the efficiency of these features in predicting gene essentiality.
• The DNN model exhibited superior prediction performance compared to other predictors evaluated in previous studies.

Conclusions:

• The developed DNN model, leveraging key genomic features, provides a highly accurate and generalizable approach for essential gene prediction in Mycobacterium tuberculosis.
• The identified key features are crucial for understanding gene essentiality and can guide future research in antimicrobial drug target identification.
• This study highlights the potential of computational methods, particularly DNNs, in advancing our understanding of bacterial genomics and facilitating the development of new treatments.