Harnessing interpretable deep learning to predict resistance in Klebsiella pneumoniae

Antimicrobial resistance poses a growing global health threat, complicating therapeutic management and increasing morbidity and mortality. Deep learning methods have emerged as effective tools for bacterial profiling using omics data, particularly in predicting antimicrobial susceptibility from genomic information. The present study focuses on identifying genomic signatures linked to resistance mechanisms using a deep learning architecture.

DeepMDC, a deep learning architecture for bacterial profiling using whole-genome data, is introduced. Due to the high cost and ambiguity of precise gene- or mutation-level annotation, phenotypic classification is formulated as a Multiple Instance Learning (MIL) problem, in which each genome is represented as a bag of instances with a single associated label. The architecture centers on a modern Hopfield network that processes all open reading frames (ORFs), including small ORFs, derived from genomic data. Interpretability is achieved through attention mechanisms, which facilitate biological insight and support hypothesis generation.

The model was evaluated against Klebsiella pneumoniae and four clinically relevant antibiotics (meropenem, cefepime, ceftazidime, and gentamicin), achieving strong performance across multiple evaluation metrics.

Notably, genes associated with resistance consistently received high attention scores during inference, validating the architecture and potentially generating new hypotheses.