RGTFormer: Predicting mutation-associated multi-drug resistance in Mycobacterium tuberculosis using a categorical...

Area of Science:

• Genomics
• Computational Biology
• Drug Resistance Studies

Background:

• Tuberculosis (TB) poses a significant global health challenge, exacerbated by rising multi-drug resistance.
• Drug resistance in TB often stems from specific mutations in target genes, necessitating early prediction for effective treatment.

Purpose of the Study:

• To develop and evaluate RGTFormer, a novel deep learning model for predicting mutation-driven drug resistance in TB.
• To assess the model's performance against existing methods using sequence and structural mutation data.

Main Methods:

• Developed RGTFormer, integrating a categorical gated transformer with a Relational Graph Convolutional Network (RGCN).
• Utilized sequence and structural features from mutations across six key anti-TB drug resistance genes.
• Employed 10-fold cross-validation and an independent test set for rigorous evaluation.

Main Results:

• RGTFormer achieved high accuracy: 98.67% on the test set and 97.15% via cross-validation.
• The model outperformed traditional machine learning and other deep learning approaches.
• Ablation studies validated the synergistic contribution of RGCN and gated attention mechanisms.

Conclusions:

• RGTFormer offers a robust, interpretable, and efficient framework for predicting TB drug resistance based on mutations.
• The model shows potential for personalized TB treatment and optimizing drug selection for resistant strains.
• Biologically interpretable predictions from RGTFormer can inform clinical decision-making.