Fig. 1

MalaiaSED accurately predicts the chromatin profiles in malaria parasites. A The general framework of the deep learning model. The input DNA sequence is encoded into the four-row matrix by one hot encoder, where each element represents the appearance of a nucleotide at a specific location. The following two convolutional layers capture the cumulative effects of short sequence patterns by analogy to motif scan. The bidirectional LSTM layer follows up to summarize long-term dependencies between captured DNA patterns. Outputs are fed into the flatten and full connection layers. We calculate the final score by dense layer with sigmoid activation. B Performances of the deep learning framework are measured by area under the curve in multiple experimental epigenetic profiles in P. falciparum and P. berghei