A new automatic method for human coding and non-coding zones characterization and classification based on FCGR coding and CNN classifier

Human genes characterization is a difficult task given the length and the enormous amount of information contained in them especially when it comes to coding and non-coding zones. So, compositional calculations and signal processing techniques will be of great help to facilitate this task. Therefore, we apply, in this work, the Frequency Chaos Game Representation (FCGR) method to characterize human exons and introns. Representing coding and non-coding regions by images allows us to classify them using a CNN model. We have reached an average value of 92.05% of accuracy rate. Then, to validate the robustness of our chosen model to input variations, we propose a new image configuration for those human exons and introns and we construct 10 new datasets containing combinations of different orders of FCG Representation. When classifying those new images by the same CNN model, we have obtained very promising results going to 89.18% of average accuracy rate.