DTL dephossite | prediction method of dephosphorylation sites based on Transfer Learning

at present , People pay little attention to dephosphorylation and its role . so far , Dephosphorylation site prediction tools are limited to a few tyrosine phosphatases . To fill this knowledge gap , This paper adopts a transfer learning strategy to develop a model based on deep learning to predict the sites that may be dephosphorylated . Based on independent test results , The model of this paper DTL-DephosSite In sensitivity （SN）, Specificity （SP） and Matthew The correlation coefficient （MCC） Excellent performance in .

Thesis link ：

https://pubmed.ncbi.nlm.nih.gov/34249915/

This paper describes a strategy that combines deep learning with transfer learning , To develop S / T and Y General dephosphorylation site prediction model of residues . The resulting model , be called DTL-DephosSite-ST and DTL-DephosSite-Y, Is the first S/T and Y Dephosphorylation site prediction model , And it has good performance . The experimental results show that , The model produced by transfer learning is significantly better than using only Bi-LSTM Developed models .

During transfer learning , Three important questions need to be answered ：（a） Transfer what ,（b） When to transfer , as well as （c） How to transfer . In order to allow the framework of this article to accommodate smaller data sets , This paper applies a two-step transfer learning scheme , It includes a pre training step and a fine-tuning step （ chart 3）. The pre training step will generate a source model , Then fine tune to fit the target data set .