We utilized 10X Genomics CellRanger software to perform cellular de-multiplexing, read alignment and Unique Molecular Identifier (UMI) quantification. Matrices containing UMI counts were filtered based on number of genes expressed per cell (>=500) and fraction of mitochondrial and ribosomal transcripts (<5%). Matrices from individual runs were combined, normalized to total UMIs per nucleus and log transformed. Genes expressed in more than five cells were considered for downstream analysis.

Dimensionality reduction was performed using truncated principle component analysis (PCA); significant principle components were selected using scree plot method. PC scores were used to perform t-Distributed Stochastic Neighbor Embedding (t-SNE), as well as to calculate nearest neighbor (nn) distances. NN edges were Jaccard distance-weighed and used to perform Louvain clustering. Clustering information was combined with t-SNE coordinates to visualize clusters.

To determine cell types constituting each cluster, a combination of expression plots for know cell type markers, regional t-SNE plots and unbiased marker discovery with Model-based Analysis of Single-cell Transcriptomics (MAST) was utilized.

To perform discovery of genes differentially expressed in ASD subjects compared to controls in each cell type, MAST was used to fit a linear mixed model and control for individual as a random effect factor and age, sex, RNA integrity number (RIN), cortical region, mitochondrial and ribosomal RNA ratios, sequencing platform, 10X capture and sequencing batch as fixed effect factors. Genes with Fold Change >=0.2 and FDR<0.01 were considered differentially expressed.