MRDPDA: A multi-Laplacian regularized deepFM model for predicting piRNA-disease associations.

Area of Science:

• Biochemistry
• Genomics
• Bioinformatics

Background:

• PIWI-interacting RNAs (piRNAs) are crucial small non-coding RNAs involved in gene regulation and genome stability.
• piRNAs show promise as biomarkers and therapeutic targets for numerous diseases.
• Current computational methods struggle to accurately predict piRNA-disease associations (PDAs) due to data limitations.

Purpose of the Study:

• To develop a novel computational method, MRDPDA, for predicting PDAs from limited, multi-source data.
• To enhance the accuracy of PDA prediction by integrating a deep factorization machine (deepFM) with specialized regularization techniques.

Main Methods:

• MRDPDA employs a deepFM model combined with regularizations from multiple limited datasets using separate Laplacians.
• A unified objective function integrates embedding loss for similarity, optimizing the embedding for prediction.
• A balanced benchmark dataset (piRPheno) and a deep autoencoder for negative set generation were utilized.

Main Results:

• MRDPDA demonstrated superior performance compared to three state-of-the-art methods on the piRPheno dataset.
• The method achieved high accuracy in both five-fold cross-validation and independent testing.
• Case studies confirmed the practical effectiveness of MRDPDA in identifying PDAs.

Conclusions:

• MRDPDA offers a robust and effective approach for predicting piRNA-disease associations, particularly with limited data.
• The developed method has significant implications for identifying novel disease biomarkers and therapeutic strategies involving piRNAs.
• This work advances computational approaches in small non-coding RNA research and disease association studies.