Mammalian piRNA target prediction using a hierarchical attention model.

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Piwi-interacting RNAs (piRNAs) are vital for genome defense in germline cells and regulate mRNA levels.
• Mammalian piRNA targeting rules remain poorly defined, hindering research into piRNA functions.

Purpose of the Study:

• To develop a robust deep learning model for predicting piRNA targets in mammals.
• To address challenges of limited data and incomplete understanding of piRNA targeting mechanisms.

Main Methods:

• Designed a deep learning model incorporating specific sub-networks based on inferred piRNA targeting patterns.
• Employed transfer learning to overcome data scarcity issues.
• Validated model performance on an independent test dataset.

Main Results:

• The deep learning model achieved high accuracy (98.5%) in predicting piRNA targets.
• The model successfully predicted targets for all available mouse and human piRNAs in the database.

Conclusions:

• A novel deep learning framework significantly improves piRNA target prediction.
• This tool overcomes data limitations and advances the understanding of piRNA function.
• The developed framework and results are publicly available for research use.