The invention provides a transcription factor binding site prediction algorithm and device across transcription factors. The method comprises the following steps: 1, predicting amino acids capable ofbinding to DNA in all transcription factors, namely DNA binding sites, wherein the predicted DNA binding sites are mainly used for measuring contributions of labeled data of different transcription factors in a target transcription factor model training process; 2, learning a representation vector of transcription factors from a sequence composed of the predicted DNA binding sites; 3, learning thehigh-order dependency relationship of the DNA fragments from the histone modification characteristics of the DNA fragments; 4, learning the low-order dependency relationship of the DNA fragments fromthe sequence characteristics of the DNA fragments; 5, splicing the learned vector representation of the transcription factors, the high-order dependency relationship and the low-order dependency relationship of the DNA fragments into a feature vector, inputting the feature vector into a multilayer perceptron to classify the target DNA fragments, and judging whether the target DNA fragments are binding sites of the target transcription factor or not.