Figure 2. Different possible uses of deep learning in genomics.

Deep learning models trained with genome annotations together with the underlying genomic sequence (in light blue) can be used for thre different applications (in light green) (1) to automatically annotate the genome of a given species and for a given cell type, (2) to determine the sequence determinants of the genome functions by identifying sequence motifs (such as position weight matrix, PWM) and the effect of sequence variants, or (3) to design artificial sequences.