Fig. 1: An overview of the DeepSEED approach.
From: Deep flanking sequence engineering for efficient promoter design using DeepSEED
a The learning procedure of DeepSEED. The model was able to automatically design the promoter sequences by means of the two stages of the learning process. b Training process of the generator and predictor. The details of the sequence construction in the three promoter design tasks are shown in Supplementary Fig. 6. c Sequence generation process in the generator and genetic algorithm. The functional regions represent the functional promoter sequence space learned by the generator. N1 and N2 represent the natural sequences from the dataset. S0, S1, S2, and S3 represent the sequences in the optimization process. d The promoter design tasks in this work. The protein binding sites, as prior expert knowledge, were used to design constitutive promoters, IPTG-inducible promoters, and Dox-inducible promoters.
