Convolutional neural networks quantify antibiotic resistance in Mycobacterium tuberculosis with diagnostic grade accuracy and predict treatment response

· Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Pharmaceuticals & Biotechnology, Clinical Care & Medical Practice, Medical Devices & Health Technology · Depth: Expert, medium

Summary

Convolutional neural networks (CNNs) have been developed to predict minimum inhibitory concentrations (MICs) for eight antibiotics from *Mycobacterium tuberculosis* complex (MTBC) gene sequences. These CNN models achieve diagnostic-grade accuracy, predicting 89% of MICs within one drug concentration doubling. The models incorporate evolutionary information, protein biochemical properties, and data augmentation for rare variants. Despite being trained on less than 52% of the World Health Organization's (WHO) MTBC drug resistance mutation catalog data, the CNNs accurately predict the effects of 97% of the catalog's graded mutations. Furthermore, in a cohort of 373 patients with rifampicin-susceptible *M. tuberculosis* infections, higher CNN-predicted rifampicin MICs correlated with unfavorable treatment outcomes, indicating the clinical relevance of subtle MIC differences below the resistance threshold.

Key takeaway

For AI Scientists and Research Scientists developing diagnostic tools for infectious diseases, this work demonstrates that integrating diverse biological data dimensions (evolutionary, biochemical) into machine learning models can achieve clinical-grade accuracy. You should consider these multi-modal data inputs and data augmentation strategies to enhance model performance, especially for predicting continuous biomedical variables like MICs, which can inform treatment decisions even below traditional resistance thresholds.

Key insights

CNNs can predict antibiotic MICs from genomic data with clinical accuracy, leveraging diverse biological features.

Principles

Method

Convolutional neural networks predict MICs from MTBC gene sequences by encoding evolutionary information, protein biochemical properties, and using data augmentation for rare variants.

In practice

Topics

Best for: AI Scientist, Research Scientist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.