DeepSEA complete datasets and code

Surveying antimicrobial resistance (AMR) is essential to track the evolution and spread of resistant genes/proteins since AMR is a major cause of death, being responsible for more deaths than HIV and malaria combined.  Alignment-based annotation tools use strict similarity (>70%) cutoffs to distinguish between potential and AMR real sequences and only annotate proteins similar to those in their databases. DeepARG and AMRFinderPlus use artificial neural networks (ANN) and Hidden Markov Models (HMM) to annotate AMR proteins with remote homology. However, DeepARG needs a pre-processing step that aligns the query data and selects the most probable proteins, although the filtering uses looser cutoffs. HMMs also depend on multi-sequence alignment (MSA) and are focused on a single AMR class. Here we present DeepSEA, an alignment-free tool fitted on antimicrobial resistant proteins (APR) and non-resistant proteins (NRP) aligned and unaligned to ARP. Our results show that DeepSEA outperforms the current multi-class AMR classifiers. Furthermore, DeepSEA’s model can cluster AMR by resistant mechanisms, showing that the model's latent variables successfully captured distinguishing features of antibiotic resistance. Our tool annotated functionally validated tetracycline destructases (TDases) and confirmed the identification of a novel TDase found by HMM.