An efficient deep learning based predictor for identifying miRNA-triggered phasiRNA loci in plant.

Area of Science:

• Plant molecular biology
• Genomics
• Bioinformatics

Background:

• Phased small interfering RNAs (phasiRNAs) are crucial plant regulatory molecules derived from miRNA-mRNA interactions.
• Identifying phasiRNA loci is vital for understanding plant growth and stress responses but is experimentally intensive.
• Computational approaches are needed to efficiently analyze high-throughput sequencing data for phasiRNA discovery.

Purpose of the Study:

• To develop an accurate and efficient computational predictor, named DIGITAL, for identifying miRNA-initiated phasiRNA loci in plants.
• To leverage deep learning techniques for high-throughput analysis of phasiRNA generation pathways.

Main Methods:

• Proposed DIGITAL, a predictor combining a multi-scale residual network and a bi-directional long-short term memory network.
• Constructed a negative dataset by random nucleotide replacement in positive phasiRNA samples.
• Evaluated the predictor on two independent test datasets with varying sequence lengths.

Main Results:

• Achieved high prediction accuracy: 98.48% on one dataset and 94.02% on another.
• Demonstrated the effectiveness, robustness, and generalization ability of the DIGITAL model.
• Confirmed the model's potential for application in miRNA target recognition across different species.

Conclusions:

• DIGITAL provides an effective computational solution for identifying plant miRNA-initiated phasiRNAs.
• The predictor's accuracy and robustness support its utility in accelerating plant research.
• The freely available source code facilitates broader application and extension of the method.