HArmonized single-cell RNA-seq Cell type Assisted Deconvolution (HASCAD).

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cell composition deconvolution (CCD) estimates cell fractions from bulk gene expression data.
• Existing CCD models often use linear regression with reference gene signatures from single-cell RNA sequencing (scRNA-seq).
• Batch effects and dropout events in scRNA-seq limit the performance of current CCD methods.

Purpose of the Study:

• To develop a deep neural network (DNN) model, HASCAD, for accurate cell composition deconvolution.
• To predict the fractions of up to 15 immune cell types from bulk RNA-seq data.
• To evaluate HASCAD's performance against established CCD methods.

Main Methods:

• Developed HASCAD, a DNN model trained on simulated bulk RNA-seq data derived from normalized scRNA-seq datasets.
• Employed a Harmony-Symphony strategy for scRNA-seq data normalization.
• Benchmarked HASCAD against CIBERSORTx and quanTIseq using simulated and human PBMC RNA-seq datasets.

Main Results:

• HASCAD demonstrated superior performance compared to CIBERSORTx and quanTIseq in analyzing bulk RNA-seq data.
• Removal of batch effects in reference scRNA-seq datasets improved CCD task performance.
• Analysis of TCGA-LIHC liver cancer data revealed associations between predicted Treg and effector CD8 T cell abundance and patient survival.

Conclusions:

• HASCAD effectively predicts immune cell composition from bulk RNA-seq data.
• The model aids in investigating immune cell heterogeneity's impact on therapeutic responses.
• HASCAD is a valuable tool for analyzing immune cell populations in bulk RNA-seq studies.