AdRoit is an accurate and robust method to infer complex transcriptome composition.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Bulk and spatial transcriptomics offer whole-transcriptome insights but lack cell type resolution.
• Existing in-silico deconvolution methods struggle with performance in complex tissue samples.
• Accurate cell composition inference is crucial for understanding tissue organization and function.

Purpose of the Study:

• To present AdRoit, a novel computational method for accurate and robust cell composition inference from mixed transcriptome data.
• To enhance the comparability of data across different sequencing techniques, including single-cell, bulk, and spatial transcriptomics.
• To provide a computationally efficient tool for deconvolution of complex biological samples.

Main Methods:

• AdRoit utilizes single-cell RNA sequencing profiles as a reference for cell type deconvolution.
• An adaptive learning approach is employed to reconcile differences between sequencing platforms.
• The method was systematically benchmarked on complex mixtures containing up to 30 cell types.

Main Results:

• AdRoit demonstrates superior sensitivity and specificity compared to existing deconvolution methods.
• The approach effectively handles cross-platform data discrepancies, improving readout comparability.
• Benchmarking confirmed AdRoit's robustness and accuracy in complex tissue deconvolution.

Conclusions:

• AdRoit provides an accurate, robust, and computationally efficient solution for inferring cell composition from transcriptome data.
• The method significantly improves the utility of bulk and spatial transcriptomics for biological discovery.
• AdRoit offers a valuable tool for researchers studying complex tissues and cellular heterogeneity.