Abstract
The accurate annotation of a protein function is important for understanding life at molecular level. Nowadays, powerful high throughput proteomics technologies provide an unprecedented understanding of the human biology and disease. These technologies are generating a deluge of protein sequences available in public databases. However, a critical challenge in making sense of these sequences is the assignment of functional roles to newly discovered proteins. The approaches proposed to address this problem use a variety of biological information, such as amino acid sequence, gene expression and protein-protein interaction. By another way, deep learning has emerged as the innovation of this last decade as it uses deep architectures to learn representations of high level entities and creates an improved functional space. In this paper, we propose an approach that proposes a deep neural network to achieve classification of oxygen binding proteins using amino acid composition for protein function prediction. Two alternatives are investigated. The first one casts the tackled problem as a multiclass classification problem and the second one as a binary classification problem. The validation of the approach is achieved using Keras platform and very promising and encouraging results that outperform other state of the art results have been obtained.