DiCleave: a deep learning model for predicting human Dicer cleavage sites.

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• MicroRNAs (miRNAs) are key gene regulators, essential for cellular processes.
• miRNA maturation involves Dicer enzyme cleavage of precursor miRNAs (pre-miRNAs).
• Existing prediction models often neglect comprehensive pre-miRNA structural information and binary classification limitations.

Purpose of the Study:

• To develop an advanced deep learning model for predicting Dicer cleavage sites in pre-miRNAs.
• To leverage autoencoders for learning secondary structure embeddings of pre-miRNAs.
• To overcome limitations of existing binary classification models and incorporate multi-class prediction.

Main Methods:

• Developed a deep learning model incorporating an autoencoder for pre-miRNA secondary structure.
• Trained the model using sequence and secondary structure information.
• Evaluated model performance through benchmarking against state-of-the-art methods.

Main Results:

• The deep learning model achieved performance comparable to ReCGBM in binary classification tasks.
• The model demonstrated superior performance in multi-class classification of Dicer cleavage sites.
• The autoencoder effectively learned secondary structure embeddings, enhancing prediction accuracy.

Conclusions:

• The proposed deep learning model offers a more versatile and practical solution for Dicer cleavage site prediction.
• Incorporating secondary structure information significantly improves prediction capabilities.
• The model's multi-class classification ability enhances its utility in understanding miRNA biogenesis.