Enhance the recognition of signal peptide domain and transmembrane domain based on transformer

In the past two decades, many algorithms have been proposed to predict the signal peptide domain and the transmembrane domain of membrane proteins. The signal peptide domain and the transmembrane domain are rich in hydrophobic Amino acids, which have similar biological properties. The prediction results of these proposed algorithms are not perfect. In this paper, transformer is used to predict the transmembrane domain and signal peptide domain of the protein. Diverse protein amino acid sequences have various biological information and diverse arrangements of amino acid residues. The encoder layer is able to concentrate on the biological properties of protein sequences as well as the context information of lengthy sequences. In the comparative experiment, our method predicts that the F1 score of the transmembrane domain would reach 0.7863. The predicted F1 score for the signal peptide domain is 0.7076, which is 0.074 higher than the comparative model’s top performance. Overall, our method outperforms other approaches in testing tasks. Accurate prediction of signal peptide domains and transmembrane domains can help researchers quickly understand the structure and biological function of proteins. Potential drug targets or inhibitor binding sites on signal peptides and transmembrane domains can be further identified by researchers, thereby designing relevant targeted drugs.