Predição de regiões promotoras em Bacillus subtilis através do uso de redes neurais artificiais e máquinas de vetor de suporte

Abstract

The transcription process concerns reading the information contained in DNA to generate the corresponding messenger RNA. To initiate the transcription process of a given gene, RNA polymerase enzyme needs to recognize the promoter region, thereby regulating gene expression. Literature proposes several computational methods to predict promoter sequences, but most of them is focused on Gramnegative bacteria. Therefore, the objective of this work is to predict promoters in intergenic regions of the Bacillus subtilis bacterium (Gram-positive) through the application of machine learning techniques: Artificial Neural Networks (RN) and Support Vector Machines (SVM). The training of the RN was performed through the Multilayer Perceptron (MLP) algorithm that is based on the error correction learning rule (backpropagation). For SVM, the kernels (maps the characteristics space to identify ideal support vectors) that stands out are Radial Basis Function (RBF) that uses a Gaussian function; SIGMOID that uses a hyperbolic tangent function; and Nu- Support Vector Classification (Nu-SVC) that limits the penalty cost between 0 and 1. The first step was to obtain the genome and the promoters recognized by the Sigma factors of Bacillus subtilis from data in public data bases. Biological data gathered from Bacillus subtilis generated 767 promoter regions, being the majority found by Sigma SigA factor. These data were processed and used as input in RN and SVM machine learning techniques. Hence, it was possible to compare the efficiency of the two solutions. In both solutions, the same input data and 5-fold cross-validation were used. We obtained 93.20% and 95.63% accuracy in the SVM application (combining the SIGMOID and RBF kernels with the Nu-SVC algorithm). With RN (MLP with 5 and 7 neurons in the hidden layer and 1 neuron in the output layer), the best results were 98.57% and 97.69% accuracy. Both results are consistent and competitive when compared to those in literature. In addition, both solutions proved the reliability of the obtained data. Finally, it is possible to state that the prediction of Bacillus subtilis promoter regions can be successfully performed both using RN and SVM, although RN has obtained better performance.