Transcriptome Complexity Disentangled: A Regulatory Molecules Approach.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Transcription factors (TFs) and microRNAs (miRNAs) are key regulators of gene expression and cellular processes.
• Understanding the minimal set of regulators needed to predict global gene expression is crucial for advancing transcriptomics.

Purpose of the Study:

• To determine if a limited number of TFs and miRNAs can accurately predict genome-wide gene expression.
• To develop predictive models for gene expression using selected regulatory molecules.
• To explore the low-dimensional structure of the transcriptome.

Main Methods:

• Analysis of 8895 cancer samples from The Cancer Genome Atlas (TCGA) across 31 cancer types.
• Unsupervised learning to identify clusters of miRNAs and TFs, selecting medoids as representative molecules.
• Development of Tissue-Agnostic and Tissue-Aware models to predict gene expression using 56 selected medoid miRNAs and TFs.

Main Results:

• Identified 28 miRNA and 28 TF clusters; medoids differentiated tissues of origin with 92.8% accuracy.
• Tissue-Aware model achieved an R2 of 0.70 in predicting gene expression, comparable to 1000 landmark genes despite using fewer molecules.
• Demonstrated that a small subset of regulatory molecules can capture the transcriptome's intrinsic low-dimensional structure.

Conclusions:

• A biology-guided approach using a minimal set of TFs and miRNAs can robustly represent transcriptome-wide gene expression.
• This method offers potential for cost-effective transcriptome assays and analysis of low-quality samples.
• Findings provide insights into gene regulation by miRNAs/TFs versus alternative mechanisms, though model transportability requires further investigation.