PAIRNet: Predicting PIWI cleavage specificity via position-aware RNA interaction modeling.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• PIWI proteins are crucial for genome integrity via piRNA-guided RNA cleavage.
• Cleave-N'-Seq (CNS-seq) maps PIWI targeting but has labor-intensive workflows.
• Systematic exploration of sequence determinants in PIWI targeting is limited.

Purpose of the Study:

• To develop PAIRNet, a deep learning framework for predicting PIWI-mediated RNA cleavage rates.
• To model guide-target interactions, considering geometry and sequence.
• To accelerate mechanistic studies of RNA-guided genome defense.

Main Methods:

• Developed PAIRNet, a deep learning framework integrating biochemical insights and computational methods.
• Encoded pairing states, mismatches, insertions, deletions, and positional embeddings.
• Employed a hybrid CNN-Transformer architecture to prioritize duplex dynamics.
• Incorporated interpretability modules (saliency maps, counterfactual analysis).

Main Results:

• PAIRNet accurately predicts PIWI-mediated RNA cleavage rates across four PIWI-guide datasets.
• Achieved significant improvements in PCC (34.7% for MILI, 14.6% for MIWI) over existing methods.
• Recapitulated key biological principles, including stringent complementarity at catalytic residues and 3' mismatch tolerance.

Conclusions:

• PAIRNet bridges biochemical precision with computational scalability for PIWI targeting analysis.
• Establishes a roadmap for designing high-specificity piRNA silencing tools.
• Accelerates mechanistic studies of RNA-guided genome defense.