Using deep neural networks and LASSO regression to predict miRNA expression changes based on mRNA data.

Area of Science:

• Multi-omics data analysis in disease research.
• Computational biology and bioinformatics.
• Gene expression profiling.

Background:

• Understanding microRNA (miRNA) and messenger RNA (mRNA) interplay is crucial for disease molecular studies.
• Paired miRNA/mRNA expression datasets are scarce, limiting integrated analysis.
• Existing miRNA-target relationship research relies on databases with experimental and computational data.

Purpose of the Study:

• To assess the feasibility of predicting miRNA expression profiles from mRNA expression data.
• To demonstrate the potential of cross-omics prediction as a proof of principle.
• To leverage publicly available mRNA data for miRNA expression inference.

Main Methods:

• Utilized artificial deep neural networks and LASSO regression for miRNA expression prediction.
• Evaluated models on seven paired miRNA/mRNA datasets from West Nile virus and HIV infections.
• Performed within-data and cross-study evaluations, exploring data augmentation and separate models for diseased/non-diseased samples.

Main Results:

• Achieved strong correlations at the individual sample level across most settings.
• Observed correlations in log-fold changes and p-values from differential expression analysis (DEA) in some datasets.
• Data augmentation consistently improved neural network performance, with less impact on LASSO models.

Conclusions:

• Cross-omics prediction of miRNA expression from mRNA data is demonstrated as feasible.
• The approach shows potential for inferring miRNA profiles where paired data is unavailable.
• Correlations in differential expression analysis suggest the utility of this method in disease research.