Benchmarking of cell type deconvolution techniques for microarray data

Cell type deconvolution algorithms have significantly advanced in recent years. However, a notable gap remains regarding the performance of these algorithms when applied to microarray data, which is widely available in numerous disease contexts. Here we compare the performance of existing cell type deconvolution algorithms across a spectrum of data preparation scenarios and diverse cell types within the context of microarray-based bulk transcriptome. Among the ten algorithms evaluated, MuSiC, qprogwc, and CIBERSORTx emerged as the top performers, demonstrating the highest correlation between the estimated and real cell type compositions ( r max = 0.74, 0.69, and 0.67, respectively). Our study provides a practical framework for microarray deconvolution, while also underscoring the necessity for greater precision in these techniques. Such methods would unlock the full potential of the vast repositories of microarray data in public domains, facilitating deeper insights into cellular compositions in various disease states, and promoting translational research.