Advances in machine learning and deep learning for antimicrobial resistance prediction

Antimicrobial resistance (AMR) is a major public health risk. Testing antimicrobial susceptibility in patients using conventional methods takes a week or more to deliver results, advances in artificial intelligence (AI), including machine learning (ML), and deep learning (DL), have created an opportunity to predict AMR with speed by combining data from different sources such as genomic, transcriptomic, proteomic, phenotypic, and clinical information. Recent advances in whole genome sequencing, metagenomic methods, and MALDI-TOF mass spectroscopy, as well as the availability of established and curated databases such as the Comprehensive Antibiotic Resistance Database (CARD), are essential components of developing robust models for predicting AMR. A large number of studies using ML and DL methods have demonstrated that they can identify resistance genes, discover new AMR determinants, predict trends in AMR, and improve antimicrobial stewardship and drug discovery. This review describes methodologies currently used to predict AMR with a focus on the types of data used, feature engineering methods, and algorithms used in pathogen-specific and pan-species studies. It discusses several current challenges associated with building predictive models including lack of diversity of datasets, class imbalance, inconsistent methods of testing, limited ability to interpret results, and the need for further validation. Future directions involve developing explainable AI models, federated learning approaches to improve datasets, and integrating digital diagnostics into electronic health records. In conclusion, ML and DL have the potential to revolutionize precision medicine, enhance AMR monitoring, and accelerate the pace of drug development.