QClus: Robust and reliable preprocessing method for human heart snRNA-seq

Abstract

AbstractSingle nuclei RNA sequencing (snRNA-seq) is widely used to study tissues and diseases. However, the technique remains challenging, as cytoplasmic RNA often contaminates nuclei-containing droplets and may even complicate the removal of empty droplets. Incomplete removal of contaminated background signal masks cell type-specific signal and interferes with differential gene expression analysis. Thus, removing empty droplets and highly contaminated nuclei is of paramount importance in snRNA-seq analysis. This is especially the case in solid tissues such as the heart. Here, we present QClus, a novel nuclei filtering method targeted to human heart samples. In the human heart, most of the contamination originates from cytoplasmic RNA, stemming from cardiomyocytes. Therefore, we use specific metrics such as splicing, mitochondrial gene expression, nuclear gene expression, and non-cardiomyocyte and cardiomyocyte marker gene expression to cluster nuclei and filter empty and highly contaminated droplets. This approach combined with other filtering steps enables for flexible, automated, and reliable cleaning of samples with varying number of nuclei, quality, and contamination levels. To confirm the robustness of our method, we ran QClus on our dataset of 32 human heart samples and compared results to six alternative methods. None of the methods resulted in the same filtering quality as QClus, each of them failing significantly in some samples. As snRNA-seq is predominantly used on unique human tissue samples, sample failure due to inadequate data processing seems unacceptable. Our study shows that, instead of using universal preprocessing methods, challenging tissue types, such as heart tissue, may benefit from methods that are more specific to the tissue under study, as QClus outperforms the other methods primarily due to improved integration of sample type-specific features.