Treegl

Treegl estimates gene regulatory networks across biological lineages using ℓ1 (Lasso) and total variation penalized linear regression to jointly model multiple, sparse networks organized on a tree genealogy.

Key Features:

• ℓ1 plus total variation penalized linear regression: Combines ℓ1 (Lasso) penalization with total variation penalties within a linear regression framework to enforce sparsity and encourage similarity and change detection along the lineage.
• Tree-genealogy modeling: Represents cell types as nodes in a branching tree genealogy and jointly infers network structures for related nodes to capture continuity and divergence of regulatory interactions.
• Information sharing for limited samples: Shares information across related networks along the lineage to enable estimation when sample sizes per cell type are small.

Scientific Applications:

• Developmental biology and differentiation studies: Models how gene regulatory networks evolve as stem cells differentiate into specialized cell types to study developmental processes.
• Cancer research: Models gene networks of treated cancer cells in relation to their malignant origins to investigate effects of drugs, progression, and potential reversion pathways.

Methodology:

Applies linear regression with ℓ1 (Lasso) and total variation penalties to jointly estimate sparsely connected gene regulatory networks across nodes of a tree genealogy, sharing information among related cell-type networks to accommodate limited sample sizes.