LinDeconSeq

LinDeconSeq deconvolutes cellular fractions from bulk gene expression data to identify cell type–specific marker genes and estimate cell-type proportions in contexts such as acute myeloid leukemia (AML).

Key Features:

• Marker Gene Identification: Uses a hybrid approach combining specificity scoring and mutual linearity strategies to identify marker genes across multiple cell types and to address limitations of pair-wise comparisons.
• Cellular Fraction Prediction: Predicts cellular fractions in bulk samples using weighted robust linear regression on the identified marker genes and demonstrates improved performance relative to MGFM and RNentropy on multiple publicly available datasets.
• Performance Metrics: Reports low average deviations (≤0.0958) and high Pearson correlations (≥0.8792) between predicted and actual cellular fractions in benchmark datasets.

Scientific Applications:

• Disease Diagnosis and Prognosis: Applied to AML to identify distinct cellular fractions such as granulocyte-monocyte progenitor (GMP), lymphoid-primed multipotent progenitor (LMPP), and monocytes (MONO) and to reveal patient heterogeneity and subgroup-specific mutation patterns.
• Clinical Outcome Prediction: Associates the GMP fraction with better prognosis and younger patient populations within a defined subgroup (SubgroupA), indicating potential use in stratifying patients by outcome.

Methodology:

LinDeconSeq combines specificity scoring and mutual linearity strategies for marker gene identification and applies weighted robust linear regression on those markers to estimate cellular fractions.