DeepDecon accurately estimates cancer cell fractions in bulk RNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Understanding cellular composition in disease-related tissues is crucial for diagnosis, prognosis, and treatment.
• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data but is costly for large studies.
• Bulk RNA sequencing is cost-effective for large-scale studies but lacks cellular resolution.

Purpose of the Study:

• To develop a computational method for deconvolving cellular composition from bulk RNA-seq data using scRNA-seq information.
• To accurately predict the fraction of cancer cells within bulk tissue samples.

Main Methods:

• Development of DeepDecon, a deep neural network model.
• Leveraging single-cell gene expression profiles to inform bulk tissue analysis.
• Implementation of an iterative refining strategy for cancer cell fraction estimation.

Main Results:

• DeepDecon accurately predicts cancer cell fractions in bulk tissues.
• The model demonstrates superior performance compared to existing decomposition methods.
• Validation on both simulated and real cancer datasets confirmed its accuracy.

Conclusions:

• DeepDecon offers a powerful tool for estimating cellular composition in bulk RNA-seq data.
• This advancement facilitates large-scale cancer studies by integrating scRNA-seq insights.
• Accurate deconvolution of tissue samples aids in disease diagnosis and treatment strategies.