Identifying piRNA targets on mRNAs in C. elegans using a deep multi-head attention network

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Piwi-interacting RNAs (piRNAs) are small non-coding RNAs crucial for silencing transposable elements and regulating endogenous transcripts.
• A systematic understanding of piRNA binding patterns and target gene identification is lacking.
• Distinctive piRNA characteristics necessitate specialized computational models for accurate target prediction.

Purpose of the Study:

• To develop the first deep learning model for identifying piRNA targeting sites on messenger RNAs (mRNAs).
• To computationally predict piRNA-mRNA binding interactions based on sequence data.
• To provide biological insights into piRNA-mRNA binding patterns.

Main Methods:

• A deep learning architecture utilizing multi-head attention was designed.
• One-hot encoding of piRNA and mRNA sequences followed by convolution and squeezing-extraction to identify motif patterns.
• A multi-head attention sub-network was incorporated to extract piRNA binding rules, simulating biological target recognition.

Main Results:

• The model achieved a high Area Under the Curve (AUC) of 93.3% on an independent test set.
• Successfully identified verified binding patterns of a synthetic piRNA.
• Demonstrated high prediction performance and suggested testable biological piRNA binding rules.

Conclusions:

• Developed the first deep learning method for identifying piRNA targeting sites on C. elegans mRNAs.
• The method exhibits high accuracy and provides biological insights into piRNA-mRNA binding.
• The piRNA binding target identification network is publicly available for download.