The DeepSTF framework in general. (A) Data pre-processing: DNA sequence and shape profiles are transformed into feature matrix. (B) Sequence clade processing: capture higher-order characteristics in multidimensional sequence profiles. (C) Shape clade processing: extract possible characteristics from multidimensional shape profiles. (D) Output processing: feature mapping is transformed into the final prediction. (E) TF Module: the improved transformer encoder structure's comprehensive processing flow.