Classification of human coding and non-coding regions based on CNN architecture

Studying role and properties related to DNA sequences is a very difficult task. This task is more difficult when it comes to human genome which is composed by coding and non-coding zones. Complexity noticed in this context is related to the fact that human DNA is made up of millions and millions of nucleotides. For that reason, it seems evident that a puissant predictive model can have an enormous advantage in proceeding the exploration of human genome. In this work, we have randomly chosen 108 human genes to represent their coding and non-coding regions by color images. We have introduced a convolutional neural network model with 8 layers to classify our images. Our model has shown very promising results. Our testing rate has exceeded 92% and our model have identified over than 95% of coding zones.