Leveraging heterogeneity across multiple datasets increases cell-mixture deconvolution accuracy and reduces...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Accurate cell proportion estimation from mixed-cell transcriptomics (deconvolution) relies on reference expression matrices (basis matrices).
• Existing basis matrices, like IRIS and LM22, may contain biological and technical biases due to limited sample diversity and platform specificity.
• These biases can impact the reliability of deconvolution results across different analytical methods.

Purpose of the Study:

• To develop and validate a novel basis matrix, immunoStates, designed to minimize biological and technical biases in cell deconvolution.
• To assess the impact of basis matrix composition on the accuracy and consistency of cell proportion estimates.
• To demonstrate the importance of incorporating sample and platform heterogeneity into basis matrices for robust deconvolution.

Main Methods:

• Construction of the immunoStates basis matrix using 6160 diverse samples across 42 microarray platforms, encompassing various disease states.
• Evaluation of existing basis matrices (IRIS, LM22) and the novel immunoStates matrix using multiple deconvolution algorithms.
• Comparative analysis of estimated cellular proportions against measured proportions to quantify accuracy and correlation.

Main Results:

• The immunoStates basis matrix significantly reduces biological and technical biases compared to existing matrices.
• The choice of deconvolution method has minimal impact on results when using a well-constructed basis matrix like immunoStates.
• Cellular proportion estimates derived using immunoStates showed higher correlation with measured proportions across all tested methods.

Conclusions:

• Basis matrices for cell deconvolution should incorporate biological and technical heterogeneity for improved accuracy.
• The immunoStates matrix offers a more robust and reliable reference for deconvolution of transcriptomics data.
• Standardizing basis matrix construction with diverse datasets is crucial for advancing the field of single-cell and bulk transcriptomics analysis.