In this work, we address the problem of genome representation and present a new model, called G2P (from Genome to Pixels), which aims to transform a genome into a two-dimensional representation in the form of an image. The intuition behind this approach is that genetic variations, as well as areas of interest in the genome, can be modeled through different levels of contrast of the pixels of an image. To demonstrate the interest of the approach, we applied the G2P model to a large amount of Mycobacterium tuberculosis genomes, distributed over 5 families. In our experiments we show, on the one hand, the different images produced. On the other hand, we apply machine learning techniques to demonstrate their predictive value while identifying the areas of the image that are useful for prediction.
Stattner et al. (Thu,) studied this question.