DTL-DephosSite: Deep Transfer Learning Based Approach to Predict Dephosphorylation Sites

Abstract

Phosphorylation, which is mediated by protein kinases and opposed by protein phosphatases, is an important post-translational modification that regulates many cellular processes, including cellular metabolism, cell migration, and cell division. Due to its essential role in cellular physiology, a great deal of attention has been devoted to identifying sites of phosphorylation on cellular proteins and understanding how modification of these sites affects their cellular functions. This has led to the development of several computational methods designed to predict sites of phosphorylation based on a protein’s primary amino acid sequence. In contrast, much less attention has been paid to dephosphorylation and its role in regulating the phosphorylation status of proteins inside cells. Indeed, to date, dephosphorylation site prediction tools have been restricted to a few tyrosine phosphatases. To fill this knowledge gap, we have employed a transfer learning strategy to develop a deep learning-based model to predict sites that are likely to be dephosphorylated. Based on independent test results, our model, which we termed DTL-DephosSite, achieved efficiency scores for phosphoserine/phosphothreonine residues of 84%, 84% and 0.68 with respect to sensitivity (SN), specificity (SP) and Matthew’s correlation coefficient (MCC). Similarly, DTL-DephosSite exhibited efficiency scores of 75%, 88% and 0.64 for phosphotyrosine residues with respect to SN, SP, and MCC.

Learning deep representations of enzyme thermal adaptation

“…We next demonstrated the application of DeepET in a transfer learning approach 35 . In transfer learning, a model pre‐trained on a large (source) data set, such as DeepET, is re‐purposed to another similar (target) problem from the same or a related domain with a smaller amount of training samples, by (a) further training and thus fine‐tuning certain layers or (b) resetting their weights and training them from scratch 36,37 . This is particularly useful for biological data sets since (a) large numbers of biological samples are expensive to collect and (b) the capacity of classical ML models like random forest are usually limited by the availability of relevant features 26 .…”

Mini-review: Recent advances in post-translational modification site prediction based on deep learning

“…Along with predicting conventional PTMs associated with functional group addition, deep learning-based methods have also been applied to predict niche-type PTMs; for instance, Chaudhari et al developed a transfer learning-based predictor (DTL-DephosSite) for dephosphorylation site prediction [127] . To collect datasets of S, T, and Y dephosphorylation sites, they integrated the experimentally verified datasets from the literature and datasets from the DEPOD database.…”

“… 2020 [122] DeepGlut Glutarylation Prokaryotes and Eukaryote CNN 10-fold CV 4,572 * https://github.com/urmisen/DeepGlut . 2020 [123] NPalmitoylDeep-PseAAC N-Palmitoylation Human DNN holdout 4,364 https://mega.nz/#F!s9cSiQIa!1jXO0NmgrhxUqOexmYuouA 2021 [124] DTL-DephosSite Dephosphorylation Human Bi-LSTM 5-fold CV and independent test 4,956 https://github.com/dukkakc/DTLDephos 2021 [127] PreCar_Deep Carbonylation Human and other Mammals CNN + BiLSTM 10-fold CV and independent test 5,003 https://github.com/QUST-SHULI/PreCar_Deep/ 2021 [125] He et al's work SUMOylation Ubiquitylation – CNN + DNN 10-fold CV 280,731 https://github.com/lijingyimm/MultiUbiSUMO 2021 [126] Note: *, Link is not working at the time of writing. Multiple, more than three species or PTM types.…”

Learning deep representations of enzyme thermal adaptation

“…We next demonstrated the application of DeepET in a transfer learning approach 30 . In transfer learning, a model pre-trained on a large dataset, such as DeepET, is re-purposed to another similar problem from the same domain with a smaller amount of training samples, by (i) further training and thus fine-tuning certain layers or (ii) resetting their weights and training them from scratch 31,32 . This is particularly useful for biological datasets since (i) large numbers of biological samples are expensive to collect and (ii) the capacity of classical machine learning models like random forest are usually limited by the availability of relevant features 26 .…”