The Simons Foundation Autism Research Initiative (SFARI) recruited a large cohort of families with autistic children who provided DNA samples and phenotypes. 54,558 families, parents and their children were sequenced, a total of 142,357 individuals with whole-exome (WES) and 12,519 with whole-genome sequencing (WGS). The data contains 32,559 trios and 8,895 quads (one sibling without autism), and 824 twins.
++The SPARK WGS August 2026 +release (SFARI Base dataset DS0000135) adds whole genomes for 45,178 individuals from +21,003 families, 20,858 of them with autism. This is a mostly new set of people: only 14 of +them are also in the 12,519-genome release and 201 in the exome release. It includes about +4,900 trios with an autistic child and 1,800 quads. The genomes were sequenced PCR-free on +the Illumina NovaSeq X at Broad Clinical Labs. Two tracks show this release. "SFARI SPARK 45k +WGS" was made from a frequency table that SFARI released with the data. It has no +autism/non-autism split, its allele counts are estimates, and variants seen only once are +left out. "SFARI SPARK 45k WGS ASD" was computed from the genotypes. It has exact counts, +the autism/non-autism split and all variants, including those seen once. For now it covers +only the DSCAM gene on chr21 (see Methods). +
+The same frequencies shown here are also available publicly on the SFARI Genome Browser. See (SPARK et al, Neuron 2018) for details.
In addition to the overall allele count (AC), allele number (AN), and allele frequency (AF), each variant record carries counts split by autism status (the asd column of the SPARK individual registration file):
Allele frequencies can also be displayed on the SFARI Genome Browser. Full CRAMs and VCFs with genotypes are available from SFARI Base. They require a data access request, which is usually reviewed quickly. More information is available in the SPARK Welcome Packet.
The genome browser track project was approved by the Simons Foundation under request -number 14584.1. The multi-sample project VCFs (pVCFs) for both the WES and WGS releases were +number 14584.1. The multi-sample project VCFs (pVCFs) for the 2024 WES and the 12,519-genome WGS releases were downloaded from SFARI Base using Globus. No minimum allele frequency cutoff was applied.
Because the genotype-level pVCFs cannot be redistributed, they were reduced to anonymous, sites-only VCFs carrying only the overall allele count (AC), allele number (AN) and frequency (AF), plus the autism-status counts described above, with the bcftools fill-tags plugin (its -S option produces the ASD/non-ASD splits), then normalized. The variants are also annotated with predicted protein consequences using bcftools csq against Ensembl gene models; that annotation is displayed in the combined frequency tracks of this collection. The exact commands for both steps are in the makeDoc file linked below.
++The WGS August 2026 release is handled differently. It was downloaded from SFARI Base with +Globus, but only the table SPARK.WGS.2026_08.gatk.pvcf_variant_frequencies.tsv was +used. SFARI made this table from the pVCFs with bcftools query. It lists chromosome, +position, REF, ALT and AF, with AF rounded to six decimals. It has no allele count or allele +number, so we estimated them. At every site, including sites on chrX and chrY, all low +frequencies are exact multiples of 1/90,356, and 90,356 is two alleles for each of the +45,178 individuals. For example, all of the 179 million alleles seen once have AF=1.1e-05. +None has 1.2e-05, which is what a site with more than about 4% missing genotypes would give. +We therefore set AN=90,356 for every variant and computed AC as AF × AN, rounded to the +nearest integer. Multiallelic records were split into one record per alternate allele. +Alleles with an estimated AC of 1 were removed. The remaining alleles were left-aligned +against the reference with bcftools norm. The script is +sparkWgs45kToVcf.sh. +
+ ++The track "SFARI SPARK 45k WGS ASD" was made from the genotype-level pVCFs of the same +release, which are split into 2.5 Mb chunks. The samples were grouped by the asd +column of the release's sample metadata file: 20,858 autistic and 24,320 non-autistic +individuals, with no sample left unassigned. As for the older SPARK tracks, bcftools ++fill-tags computed AC, AN and AF overall and per group from the called genotypes. +Missing genotypes do not count towards AN. The genotypes were then removed. Multiallelic +records were split and left-aligned with bcftools norm. Alleles that no individual +carries after joint genotyping (AC=0) were removed. There is no allele count cutoff. The +INFO field VARLEN is the length of ALT minus the length of REF: positive for +insertions, negative for deletions, 0 for substitutions. Unlike the other SPARK tracks, +this track also shows the predicted protein consequences (BCSQ) on its details +page, computed with bcftools csq against Ensembl release 115. The scripts are +sparkWgs45kPvcfToSites.sh and sparkWgs45kPvcfRange.sh. +
+The sequencing and variant-calling methods are documented as follows by SFARI:
The complete, runnable command history for downloading, counting and annotating the SFARI data, alongside every other cohort in this collection, is in the makeDoc file; search it for "SFARI SPARK". The scripts it calls, including -sparkMergeVcfAddCounts.sh (allele counts) and mergeAndAnnotate.sh (merge plus +sparkMergeVcfAddCounts.sh (allele counts), sparkWgs45kToVcf.sh (45k WGS frequency table conversion), +sparkWgs45kPvcfToSites.sh (45k WGS genotype counts) and mergeAndAnnotate.sh (merge plus consequence annotation), are in the varFreqs scripts directory.
SPARK Consortium. Electronic address: pfeliciano@simonsfoundation.org, SPARK Consortium. SPARK: A US Cohort of 50,000 Families to Accelerate Autism Research. Neuron. 2018 Feb 7;97(3):488-493. PMID: 29420931; PMC: PMC7444276