Fig. 2

Designing of DeepShape and the shape model. a pipeline of DeepShape-prime. After a uniform initiation of ribosome abundance normalized by CDS length, in each iteration, reads are mapped proportional to normalized ribosome abundance, and then a new ribosome abundance is updated. b pipeline of DeepShape. A shape model is introduced, and in each iteration, reads are mapped proportional to ribosome abundance and an instructive ribosome shape together. Then the shape model is fitted to generate a new instructive ribosome shape which can introduce heterogeneity because of different contexts. c structure of the shape model. The model contains an embedding layer which converts the codons to ‘one-hot’ vectors, three convolutional layers and a full connection layer