What question did this study set out to answer?

To improve the prediction accuracy of lysine methylation sites using deep learning techniques and multitask learning.

February 26, 2026Scientific Reports0 citationsOpen Access

Leveraging learned representations and multitask learning for lysine methylation site discovery

FCFrançois CharihConcordia University MBMullen BoulterCarleton University KBKyle K. BiggarOttawa Allergy Research Corporation

Key Points

To improve the prediction accuracy of lysine methylation sites using deep learning techniques and multitask learning.
Developed a transformer-based model named MethylSight 2.0 for predicting lysine methylation sites.
Integrated information from other post-translational modifications using multitask learning.
Used parallel reaction monitoring mass spectrometry to validate the model.”],
: [
,
,
,
iolog size of approximately 155,000 methylation sites, with expectations of detecting ~47,000 novel sites.
MethylSight 2.0 achieves state-of-the-art accuracy for predicting lysine methylation sites.
Identified 68 novel lysine methylation sites through validation experiments.
Expected to detect substantially more methylation sites compared to competing methods.

Abstract

Lysine methylation is a dynamic and reversible post-translational modification of proteins carried out by lysine methyltransferase enzymes. The role of this modification in epigenetics and gene regulation is relatively well understood, but our understanding of the extent and the role of lysine methylation of non-histone substrates remains somewhat limited. Several lysine methyltransferases which methylate non-histone substrates are overexpressed in a number of cancers and are believed to be key drivers of cancer progression. There is great incentive to identify the lysine methylome, as this is a key step in identifying drug targets. While numerous computational models have been developed in the last decade to identify novel lysine methylation sites, the accuracy of these models has been modest, leaving much room for improvement. In this work, we leverage the most recent advancements in deep learning and present a transformer-based model for lysine methylation site prediction which achieves state-of-the-art accuracy. In addition, we show that other post-translational modifications of lysine are informative and that multitask learning is an effective way to integrate this prior knowledge into our lysine methylation site predictor, MethylSight 2.0. Finally, we validate our model by means of parallel reaction monitoring mass spectrometry experiments and identify 68 novel lysine methylation sites. This work constitutes another contribution towards the completion of a comprehensive map of the lysine methylome by providing a revised estimate of its extent to approximately 155,000 sites. Of those, MethylSight 2.0 is expected to correctly detect ~ 47,000, which is substantially more than expected with competing methods, which we show to be less sensitive on a subset of experimentally validated novel methylation sites. We foresee that MethylSight 2.0, whose performance significantly surpasses that of competing models, will facilitate the discovery of a large number of novel methylation sites.

AIに質問

Bookmark

View Full Paper