A deconvolution method and its application in analyzing the cellular fractions in acute myeloid leukemia samples

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Identifying cell type-specific genes (markers) is crucial for analyzing bulk sample gene expression data.
• Pair-wise comparisons often fail to identify markers specific across multiple conditions.

Purpose of the Study:

• To develop a hybrid tool, LinDeconSeq, for accurate identification of marker genes and prediction of cellular fractions.
• To assess the utility of LinDeconSeq in disease diagnosis and prognosis.

Main Methods:

• LinDeconSeq employs specificity scoring and mutual linearity for marker gene identification across multiple cell types.
• Weighted robust linear regression is used for predicting cellular fractions based on identified markers.

Main Results:

• LinDeconSeq demonstrated superior accuracy and reproducibility in marker identification compared to existing methods.
• The tool achieved low average deviations and high Pearson correlations in predicting cellular fractions on benchmark datasets.
• Predicted cellular fractions, particularly granulocyte-monocyte progenitor (GMP), were associated with acute myeloid leukemia (AML) diagnosis, prognosis, and patient subgroups.

Conclusions:

• LinDeconSeq is a freely available, open-source tool for marker identification and deconvolution.
• The tool offers comparable accuracy to current methods and has broad applications in clinical outcomes and disease mechanisms.