SnapKin: a snapshot deep learning ensemble for kinase-substrate prediction from phosphoproteomics data

Mass spectrometry (MS)-based phosphoproteomics enables the quantification of proteome-wide phosphorylation in cells and tissues. A major challenge in MS-based phosphoproteomics lies in identifying the substrates of kinases, as currently only a small fraction of substrates identified can be confidently linked with a known kinase. By leveraging large-scale phosphoproteomics data, machine learning has become an increasingly popular approach for computationally predicting substrates of kinases. However, the small number of high-quality experimentally validated kinase substrates (true positive) and the high data noise in many phosphoproteomics datasets together impact the performance of existing approaches. Here, we aim to develop advanced kinase-substrate prediction methods to address these challenges. Using a collection of seven large phosphoproteomics datasets, including six published datasets and a new muscle differentiation dataset, and both traditional and deep learning models, we first demonstrate that a pseudo-positive learning strategy for alleviating small sample size is effective at improving model predictive performance. We next show that a data re-sampling based ensemble learning strategy is useful for improving model stability while further enhancing prediction. Lastly, we introduce an ensemble deep learning model ( SnapKin) incorporating the above two learning strategies into a snapshot ensemble learning algorithm. We demonstrate that the SnapKin model achieves overall the best performance in kinase-substrate prediction. Together, we propose SnapKin as a promising approach for predicting substrates of kinases from large-scale phosphoproteomics data. SnapKin is freely available at https://github.com/PYangLab/SnapKin.