We present a novel Deep Learning method for the Unsupervised Clustering of DNA Sequences (DeLUCS) that does not require sequence alignment, sequence homology, or (taxonomic) identifiers. DeLUCS uses Frequency Chaos Game Representations (FCGR) of primary DNA sequences, and generates “mimic” sequence FCGRs to self-learn data patterns (genomic signatures) through the optimization of multiple neural networks. A majority voting scheme is then used to determine the final cluster assignment for each sequence. The clusters learned by DeLUCS match true taxonomic groups for large and diverse datasets, with accuracies ranging from 77% to 100%: 2,500 complete vertebrate mitochondrial genomes, at taxonomic levels from sub-phylum to genera; 3,200 randomly selected 400 kbp-long bacterial genome segments, into clusters corresponding to bacterial families; three viral genome and gene datasets, averaging 1,300 sequences each, into clusters corresponding to virus subtypes. DeLUCS significantly outperforms two classic clustering methods (K-means++ and Gaussian Mixture Models) for unlabelled data, by as much as 47%. DeLUCS is highly effective, it is able to cluster datasets of unlabelled primary DNA sequences totalling over 1 billion bp of data, and it bypasses common limitations to classification resulting from the lack of sequence homology, variation in sequence length, and the absence or instability of sequence annotations and taxonomic identifiers. Thus, DeLUCS offers fast and accurate DNA sequence clustering for previously intractable datasets.

This study presents a machine learning-based analysis supporting the hypothesis that microbial adapta- tions to extreme temperatures and pH conditions can result in a pervasive environmental component within their genomic signatures. To this end, an alignment-free method was used in conjunction with both supervised and unsupervised machine learning, to analyze genomic signatures extracted from a curated dataset of approximately 700 extremophilic (temperature, pH) bacteria and archaea genomes. The dataset analyzed in this paper is, to the best of our knowledge, the largest and most comprehensive to date for the study of the genomic signatures of extremophilic organisms. The supervised learning identified specific sets of k-mers that are relevant for the identification of the environmental components in the genomic signatures of extremophiles. For k = 3, our findings are consistent with amino acid compositional biases and codon usage patterns in coding regions that were previously attributed to extreme environment adaptations. The unsupervised learning of unlabelled sequences, by multiple algorithms, identified some hyperthermophilic organisms with large genomic signature similarities in spite of belonging to different domains This computational finding was confirmed by supervised learning experiments under challenging training scenarios, and corroborated by the presence of shared phenotypic traits and characteristics of the isolating environments.