BITSER: uma ferramenta para classificação de genomas virais

Abstract

The exponential increase of available biological data, augmented by high-throughput sequen- cing technologies, has led to a particular case of big data research known as bioinformatics. Therefore, there is a need to develop methods for analyzing biological data that are efficient, interpretable and data-driven for knowledge discovery. This paper proposes the BITSER method as a new alignment-free approach for feature extraction and classification of biological sequen- ces. BITSER makes use of adaptations of texture analysis techniques from the field of computer vision, such as Local Binary Pattern (LBP) and derived aspects to extract sequence histograms, based on the electron-ion interaction potential (EIIP) values of nucleotides. The method adopts files in the FASTA format as input data, with no need for alignment or prior annotation, and offers a direct representation of characteristics with biological significance. The validation of BITSER was performed considering the SARS-CoV-2 and DENV viral genomes, achieving classification accuracy values of over 99% in various classification algorithms. The feature importance analy- sis enabled a significant dimensional reduction of features, preserving the discriminatory power of the method with only a few features per data set. Compared to existing methods, BITSER indicated superior results, computational efficiency, and interpretability. These results reinforce the method's robustness, scalability and potential for large-scale application in genomic analysis.