Gramep: método livre de alinhamento baseado no princípio de máxima entropia para a identificação de mutações snps e classificação de sequências biológicas

Abstract

With advances in DNA sequencing technology, the volume of available genomic data is becoming increasingly large and complex. Handling this vast amount of data presents a challenge in pursuing more robust and efficient analyses. The analysis of genomic variations traditionally employs alignment techniques, which can be computationally expensive and, in some scenarios, arbitrary, making them unfeasible in contexts with a high data load to be analyzed simultaneously. One alternative that is gaining attention is the development of alignment-free methods, which allow for a more efficient and scalable analysis of large volumes of genomic data. In this regard, this study proposes a tool capable of identifying single nucleotide polymorphisms (SNPs) in DNA sequences. The method begins by estimating the frequency of k-mer occurrences in all input sequences and, using the principle of maximum entropy, selects only the most informative k-mers for the variant under analysis, resulting in identifying the most informative and unique k-mers concerning the reference sequence. Each unique k-mer is then used to determine mutations relative to the reference by employing the Levenshtein distance. In this way, it is possible to extract information about which SNP mutation has occurred, its location, and its frequency of occurrence. Furthermore, the method can be applied to the classification of new sequences, as the occurrence frequencies of the unique k-mers for the considered variants are used as features in a machine-learning model. This enables the prediction of the class of new sequences. Results demonstrated that the method is capable of analyzing thousands of viral genome sequences simultaneously within a feasible computational time, obtaining information on SNPs in all regions of the analyzed sequences. Regarding classification, a comparison with an existing method in the literature shows satisfactory results, with superior performance metrics and lower computational resource consumption. The method was implemented in Python and Rust and is openly available to the scientific community.