Missing cell types in single-cell references impact deconvolution of bulk data but are detectable.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• RNA sequencing advances enable gene expression studies in bulk and single cells.
• Cell type proportion analysis via bulk data deconvolution relies on complete single-cell references.
• Missing cell types in single-cell references present a challenge for deconvolution accuracy.

Purpose of the Study:

• To investigate the impact of missing cell types on RNA sequencing deconvolution methods.
• To evaluate the performance of deconvolution algorithms under scenarios with absent cell types.
• To explore methods for recovering missing cell-type information.

Main Methods:

• Simulated bulk RNA sequencing data using paired single-cell and single-nucleus RNA sequencing data.
• Evaluation of three deconvolution methods under varying missing cell type conditions.
• Application of non-negative matrix factorization to recover missing cell-type profiles from residuals.
• Analysis of real bulk RNA sequencing data from cancerous and non-cancerous samples.

Main Results:

• Deconvolution performance is significantly influenced by the number and similarity of missing cell types.
• Single-nucleus RNA sequencing effectively captures cell types absent in single-cell counterparts.
• Missing cell-type expression patterns can be identified and recovered from residuals.
• Real data analysis corroborated simulation findings, with residual patterns indicating likely missing cell types.

Conclusions:

• Current deconvolution methods should be adapted to account for the potential absence of cell types in reference datasets.
• The recovery of missing cell-type information from residuals offers a promising avenue for improving deconvolution accuracy.
• Findings provide a foundation for developing more robust deconvolution tools.