GLaD

GLaD performs mixed-membership classification of genomic data to learn sparse subtype-specific biomarker signatures and estimate subtype distributions within solid tumor samples for analysis of genetic heterogeneity.

Key Features:

• Mixed-Membership Framework: Models mixed membership to represent multiple tumor subtypes within a single sample.
• Sparse Biomarker Signatures: Learns sparse biomarker signatures specific to each subtype.
• Subtype Distribution Estimation: Estimates a distribution over possible subtypes for each sample to quantify subtype proportions.
• Modeling of Genetic Heterogeneity: Accounts for genetic heterogeneity within and between solid tumors during classification.

Scientific Applications:

• Genomic Heterogeneity Analysis: Classifies tumors by genomic biomarkers to characterize genetic heterogeneity in solid cancers.
• Clinical Sample Evaluation (TCGA): Applies to clinical cohorts such as the Cancer Genome Atlas (TCGA) where many samples are mixtures of multiple subtypes.
• Validation in Simulated and In-Vitro Mixture Experiments: Validated on simulated data and in-vitro mixture experiments to assess performance on controlled mixtures.

Methodology:

Computational model integrates genomic data to jointly learn subtype-specific sparse biomarkers and sample-level subtype distributions; implemented as a Python module.