miRbiom

miRbiom predicts microRNA (miRNA) expression profiles from RNA-seq data by modeling RNA binding protein (RBP)–miRNA interactions with Bayesian causal networks and an XGBoost machine learning framework.

Key Features:

• Bayesian Causal Networks: Models conditional dependencies among RBPs and miRNAs to capture spatio-temporal dynamics of miRNA biogenesis beyond Drosha/Dicer-centric models.
• Machine Learning Integration (XGBoost): Embeds Bayesian network-derived features into an XGBoost framework to quantitatively predict miRNA formation levels from component expression data.
• Data sources and scale: Developed using diverse datasets including CLIP-seq, RNA-seq, and miRNA-seq totaling over 25 terabytes.
• Predictive accuracy: Achieves an average accuracy of 91% validated against a large corpus of experimentally established data.
• Direct miRNA expression prediction: Predicts expression levels for 1,204 human miRNAs directly from RNA-seq without requiring separate miRNA-seq or array data.

Scientific Applications:

• Regulatory research: Enables inference of miRNA regulatory profiles to study post-transcriptional gene regulation mediated by RBPs and miRNAs.
• Disease mechanism analysis: Facilitates investigation of miRNA involvement in disease mechanisms by providing predicted miRNA expression across conditions.
• Therapeutic research: Supports evaluation of miRNA-related therapeutic interventions and downstream effects through predicted miRNA profiles.

Methodology:

Bayesian causal networks were constructed from extensive datasets (CLIP-seq, RNA-seq, miRNA-seq) to model RBP–miRNA interactions, these networks were integrated into an XGBoost machine learning system to predict miRNA formation levels from component expression, and the predictive models were validated against experimental data.