Extended Data Figure 1: Methodology, quality control, and differential expression replication analysis.

a, RNA-seq workflow (see Supplementary Information for details). b, RNA-seq quality and alignment statistics from this study, including RNA integrity number (RIN), sequencing depth (aligned reads), proportion of reads mapping to different genomic regions, and bias in coverage from the 5′ to the 3′ ends of transcripts. c, RNA-seq read coverage relative to normalized gene length across transcript length across samples. d, Dependence between coverage and RIN across gene body. e–g, Correlation of transcript model quantifications comparing the union exon model (used throughout this study), the whole gene model (which includes introns), and the Cufflinks approach⁴³ to estimating FPKM. h, Summary table describing the characteristics of the matched covariate data used in the DGE and differential alternative splicing (DS) analysis of ASD in cortex and cerebellum. This includes the number of samples overlapping with our previous work⁸, the age and RIN distributions, and the dependence between diagnosis and age and RIN (summarized from Supplementary Table 1). i, Independent replication of ASD versus control DGE fold changes between previously evaluated and new ASD samples in cortex by RNA-seq using samples from ref. 8 (similar to Fig. 1a, but with RNA-seq in all samples). j, Correlation of P value rankings with Spearman’s correlation across different DGE methods for DGE analysis in cortex, comparing the ‘full model’ (LME P value) described in the Supplementary Information with other methods. Methods include removal of three additional principal components of sequencing surrogate variables(SVs) (LME with 5 SVs, top left), application of a permutation analysis for DGE P value computation (LME P, permuted, top right), application of variance-weighted linear regression for DGE⁴⁴ (limma voom, middle left), application of surrogate variable analysis for DGE⁴⁵ (full model + 17 SVs, middle right), and application of DESeq2 with the full model⁴⁶, which uses a negative binomial distribution (bottom left). k. Comparison of fold changes between frontal cortex (FC) and temporal cortex (TC) for all samples, demonstrating similar changes in both regions. l, Average linkage hierarchical clustering of samples in ASD cortex using the top 100 upregulated and top 100 downregulated protein coding genes, demonstrating that confounders do not drive clustering of about two-thirds of samples. m, The first principal component of the cortex DGE set is primarily associated with diagnosis, and not with other factors. The red line marks a Bonferroni-corrected P = 0.05.