miRSCAPE

miRSCAPE infers micro-RNA (miRNA) expression levels from RNA sequencing (RNA-seq) profiles using a machine learning framework to enable investigation of miRNA activity in bulk and single-cell transcriptomic data.

Key Features:

• Machine Learning Framework: Predicts miRNA expression from bulk RNA-seq profiles using a machine learning approach.
• Validation Across Ten Tissues: Validated across ten different tissues using approximately 10,000 tumor and normal bulk samples.
• Cell Type-Specific Inference: Achieves a predicted versus observed fold-difference correlation of approximately 0.81 in two independent datasets (HEK-GBM, Kidney-Breast-Skin).
• Cross-Species Application: When trained on human hematopoietic cancers, identifies active miRNAs in mouse hematopoietic cell lines with auROC = 0.67.
• Single-Cell RNA-seq Integration: Applied to infer miRNA activities in single-cell RNA-seq clusters from Pancreatic and Lung cancers and across 56 cell types in the Human Cell Landscape (HCL).

Scientific Applications:

• Gene Regulatory Network Analysis: Provides inferred miRNA expression to support reconstruction and interpretation of gene regulatory networks.
• Cancer Research: Enables investigation of miRNA activity differences between tumor and normal samples across multiple tissues.
• Single-Cell miRNA Profiling: Supports inference of cell type–specific miRNA activities from single-cell RNA-seq cluster profiles.
• Cross-Species Comparative Studies: Facilitates identification of conserved or active miRNAs across species using models trained on human data.
• Development and Homeostasis Studies: Allows examination of miRNA roles in development and homeostasis through inferred expression patterns.

Methodology:

Uses a machine learning model trained on bulk RNA-seq data (including ~10,000 tumor and normal samples across ten tissues and human hematopoietic cancer datasets) to predict miRNA expression, with performance assessed via predicted versus observed fold-difference correlation and auROC metrics and applied to single-cell RNA-seq clusters.