DNA protein binding motif prediction based on fusion of expectation pooling and LSTM

Abstract

During the process of DNA being expressed by transcription factors and ultimately generating proteins, motifs are used to label DNA sequences for transcription factors. Thus to predict DNA protein interaction is essentially a task to determine the DNA binding motif. This paper combined CNN and LSTM to the classification and prediction of DNA binding motifs. Experimental results proved that, compared with the classical CNN model, the CNN-LSTM fusion model can achieve a higher prediction accuracy for DNA motifs, for the ACC, ROC and other indicators of the latter are better than the former. Further, expectation pooling method was added to improve the recognition accuracy, which provides a feasible idea for the prediction of DNA binding motifs.