A deep learning method to integrate extracelluar miRNA with mRNA for cancer studies

Area of Science:

• Biotechnology
• Bioinformatics
• Genomics

Background:

• Extracellular miRNAs (exmiRs) and intracellular mRNAs are crucial biomarkers and therapeutic targets.
• Current methods for exmiR data suffer from noise, and mRNA analysis requires invasive procedures.
• Improved exmiR data quality and noninvasive mRNA assessment are essential for disease research.

Purpose of the Study:

• To develop a novel deep-learning model for cross-predicting exmiRs and mRNAs.
• To integrate exmiR and mRNA data for enhanced biomarker discovery.
• To enable noninvasive assessment of intracellular mRNA expression from exmiR data.

Main Methods:

• Developed CrossPred, a deep-learning multi-encoder model utilizing contrastive learning.
• Created a shared embedding space to integrate exmiR and mRNA data.
• Applied the model to predict mRNA from exmiR data and vice versa, evaluated on cancer datasets.

Main Results:

• CrossPred demonstrated superior performance compared to baseline models (encoder-decoder, exmiR/mRNA-specific, VAE).
• The model successfully predicted intracellular mRNA expression from noisy exmiR data.
• Identified key exmiRs and mRNAs associated with cancer, revealing their bidirectional relationship.

Conclusions:

• CrossPred offers a powerful tool for integrating exmiR and mRNA data.
• The study provides new insights into the complex interplay between exmiRs and mRNAs in disease.
• This approach facilitates noninvasive biomarker discovery and disease mechanism investigation.