Predicting antimicrobial mechanism-of-action from transcriptomes: A generalizable explainable artificial intelligence approach | JoVE Visualize

Area of Science:

Biomedical Science
Computational Biology
Antimicrobial Research

Background:

Antibiotic resistance is a growing global health threat, necessitating the discovery of new compounds with novel mechanisms of action (MOA).
Current methods for identifying novel antibiotics are often labor-intensive and low-throughput, hindering rapid progress.
There is a critical need for advanced computational approaches to accelerate the identification and characterization of potential antibiotic candidates.

Purpose of the Study:

To develop and validate an explainable artificial intelligence (AI) classification methodology for predicting the MOA of compounds.
To enhance the throughput and reduce the labor involved in identifying novel antibiotics.
To provide biological insights into the concept of MOA and antibiotic diversity.

Main Methods:

Development of a Hierarchical Ensemble of Classifiers (CoHEC) model, incorporating a novel feature selection algorithm named Clairvoyance.
Evaluation of the CoHEC model using whole transcriptome responses from Escherichia coli challenged with known antibiotics and crude extracts.
Validation on independent datasets, including transcriptomics data from a different E. coli strain and metabolomics data from Mycobacterium smegmatis.

Main Results:

The CoHEC model accurately predicted the primary MOA of unobserved compounds (purified and crude extracts) with >99% accuracy.
The model correctly identified darobactin, a newly discovered antibiotic, as possessing a novel MOA.
The methodology demonstrated high performance on diverse datasets, outperforming previous metrics and providing biological interpretability.

Conclusions:

The developed explainable AI methodology offers a high-throughput, accurate, and interpretable approach for antibiotic MOA prediction.
This tool can significantly accelerate the discovery pipeline for novel antibiotics crucial in combating antimicrobial resistance.
The study suggests that MOA is a nuanced concept, implying vast undiscovered antibiotic diversity within known MOA categories.