SUBSTRA: Supervised Bayesian Patient Stratification.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Precision medicine relies on effective patient stratification.
• Transcriptional data offers a rich source for identifying patient subgroups.
• Existing methods often compromise between predictive accuracy and feature interpretability.

Purpose of the Study:

• To develop a novel patient stratification method balancing predictive performance and interpretability.
• To identify patient subtypes and biologically relevant transcript clusters from transcriptional data.
• To apply the method to diverse clinical settings for improved patient subgrouping.

Main Methods:

• Introduced SUBSTRA, a Bayesian method employing regularized biclustering.
• Iteratively re-weights feature importance for optimized phenotype prediction.
• Identifies interpretable subtype-specific transcript clusters.

Main Results:

• SUBSTRA successfully identifies phenotype-relevant patient subtypes and transcript clusters.
• The method achieves predictive performance competitive with supervised approaches.
• Demonstrated utility in drug response prediction, cancer diagnosis, and transplant rejection.

Conclusions:

• SUBSTRA offers a robust approach for patient stratification using transcriptional data.
• The method effectively balances predictive power with biological interpretability.
• Provides a valuable tool for advancing precision medicine initiatives.