Classificação de proteínas expostas na superficie com Random Forest

Abstract

One of the leading causes of disease and death worldwide is bacterial infections. For example, we can mention tuberculosis, caused by the bacterium Mycobacterium tuberculosis, which still kills one and a half million people worldwide yearly, according to recent data from the World Health Organization. Vaccination is the best strategy to combat these infections. However, developing vaccines for pathogens have obstacles, such as identifying target proteins. For Mycobacterium tuberculosis alone, there are more than 4,000 proteins that are candidates for targets for building a vaccine. An alternative is the use of genomic information in the search for proteins that are good candidates. This work aimed to test popular Machine Learning algorithms, implemented in WEKA software, to classify surface exposed proteins, using the hydrophobic pattern in the amino acid sequence of 40 genomes causing severe human diseases. The random forests had a 72.83% accuracy, performing as well as the Support Vector Machine and Multilayer Perceptron algorithms, reaching of 70 and 65% accuracy, respectively. The results show the satisfactory performance of the algorithms used to classify proteins exposed on the surface, especially considering the difficulty in identifying them.