MT-EpiPred: Multitask Learning for Prediction of Small-Molecule Epigenetic Modulators

Area of Science:

• Biochemistry
• Pharmacology
• Computational Biology

Background:

• Epigenetic modulators are vital for disease treatment.
• Understanding their network-wide effects is crucial, not just individual targets.

Purpose of the Study:

• Introduce MT-EpiPred, a multitask learning method for predicting compound activity against 78 epigenetic targets.
• Evaluate MT-EpiPred's performance and compare it to existing methods.

Main Methods:

• Developed MT-EpiPred, a multitask learning model.
• Trained and validated the model on a dataset of compounds and epigenetic targets.
• Applied MT-EpiPred to predict the target of a novel compound and validated findings in vitro.

Main Results:

• MT-EpiPred achieved an average auROC of 0.915.
• The model demonstrated proficiency in handling few-shot targets.
• Identified KDM4D as a potential target for a novel compound, with in vitro validation (IC50 = 4.8 μM).

Conclusions:

• MT-EpiPred offers superior predictive performance and a broader target scope than existing methods.
• The web server provides accessible and accurate tool for discovering epigenetic modulators.
• Facilitates the development of selective inhibitors and network-level impact assessment.