Feature Importance Analysis of Non-coding DNA/RNA Sequences Based on Machine Learning Approaches

Autor(es):Mostrar menos -

Silva de Almeida, Breno Livio ; Queiroz, Alvaro Pedroso ; Avila Santos, Anderson Paulo ; Bonidia, Robson Parmezan ; da Rocha, Ulisses Nunes ; Sanches, Danilo Sipoli ; de Leon Ferreira de Carvalho, Andre Carlos Ponce ; Stadler, PF ; Walter, MEMT ; Hernandez-Rosales, M ; Brigido, MM

• Tipo de documento: Artigo Científico
• Fonte: ADVANCES IN BIOINFORMATICS AND COMPUTATIONAL BIOLOGY, BSB 2021; v. 13063, p. 12-pg., 2021-01-01.

Resumo

Non-coding sequences have been gained increasing space in scientific areas related to bioinformatics, due to essential roles played in different biological processes. Elucidating the function of these noncoding regions is a relevant challenge, which has been addressed by several Machine Learning (ML) studies in various fields of ncRNA, e.g., small non-coding RNAs (sRNAs) and Circular RNAs (circRNAs). The identification of these biological sequences is possible through feature engineering techniques, which can help point out specifics in different types of problems with ML. Thereby, there are recent studies focusing on interpretable computational methods, i.e., the best features based on feature importance analysis. For that reason, in this study we have proposed to explore different features descriptors and the degree of importance involved for classification task, using two case studies: (1) prediction of sRNAs in Bacteria and (2) prediction of circRNA in Humans. We developed a general pipeline using hybrid feature vectors with mathematical and conventional descriptors. In addition, these vectors were generated with MathFeature package and feature selection techniques in both case studies. Finally, our experiments results reported high predictive performance and the relevance of combining conventional and mathematical descriptors in different organisms. (AU)

• Processo FAPESP: 21/08561-8 - Classificação e análise de sequências bacterianas de RNA não-codificante com a utilização de técnicas de aprendizado de máquina
• Beneficiário: Breno Livio Silva de Almeida
• Modalidade de apoio: Bolsas no Brasil - Iniciação Científica