SDePER: a hybrid machine learning and regression method for cell-type deconvolution of spatial barcoding-based transcriptomic data

Spatial barcoding-based transcriptomic (ST) data require cell type deconvolution for cellular-level downstream analysis. Here we present SDePER, a hybrid machine learning and regression method, to deconvolve ST data using reference single-cell RNA sequencing (scRNA-seq) data. SDePER removes the systematic difference between the ST and scRNA-seq data (platform effects) explicitly and efficiently to ensure the linear relationship between ST data and cell type-specific expression profile. It also considers sparsity of cell types per capture spot and across-spots spatial correlation in cell type compositions. Based on the estimations, SDePER imputes for cell type compositions and gene expression at enhanced resolution. We assessed the performance of SDePER and six existing methods using simulations and four real datasets. All results showed that SDePER achieved significantly more accurate and robust results than the existing methods suggesting the importance of considering platform effects, sparsity and spatial correlation in cell type deconvolution. We measured the ST data of a human IPF lung using the 10x Genomics Visium platform to identify the transcriptome of different lung compartments including the bronchi, vascular, mesenchyme and immune compartment. Samples were obtained from explanted lung. We selected one block of frozen lung tissue obtained from a patient with Idiopathic Pulmonary Fibrosis (IPF). Sections of 10 μm (fresh frozen samples) were cut from the blocks onto Visium slides (10X Genomics) and processed according to the manufacturer’s protocol Tissue sections were Hematoxylin and Eosin stained and finally imaged (20X) using a scanning microscope (EvosM700, ThermoFischer Scientific). Tissue was permeabilized and mRNAs were hybridized to the barcoded capture probes directly underneath. cDNA synthesis connects the spatial barcode and the captured mRNA. After RT and amplification by PCR, Dual-indexed libraries were prepared as in the 10X Genomics protocol, and sequenced (2 samples/ HiSeq 6000 flow cell) with read lengths 28 bp R1, 10 bp i7 index, 10 bp i5 index, 90 bp R2.