Nearly 1 million genomes reveal what an average human genome contains

A whole-genome survey of 879,461 adults from the UK Biobank and All of Us found that a typical human genome contains approximately 4.3 million to 5.3 million small variants, alongside dozens of large copy-number changes, around 63 de novo mutations and a small number of variants associated with rare inherited disease. The analysis used short-read sequencing data from two independent population biobanks and provides individual-level reference ranges for common, rare, coding, non-coding and disease-associated variation across six genetically inferred ancestry groups.

Most of the variation in an individual genome is common and non-coding. Based on within-cohort ancestry-group allele frequencies, a median of 98.4% of the single-nucleotide variants and short insertions or deletions in each genome had a minor allele frequency above 1%. Only 1.34% were rare, with frequencies between 0.01% and 1%, and 0.26% were ultrarare, below 0.01%. When the most severe predicted consequence was assigned across Gencode version 49 transcripts, 38.3% of these small variants occurred in protein-coding genes, including 0.54% in coding exons and 1.2% in untranslated regions. Another 26.7% occurred in non-protein-coding RNA genes, while about 35% lay outside annotated protein or RNA genes and had no predicted functional consequence.

The analysis also quantified variants with possible effects on protein sequence or splicing. In coding exons of MANE Select transcripts, the median genome contained about 9,600 missense variants, of which 28 had a REVEL score above 0.7, indicating a higher likelihood of pathogenicity. It contained roughly 11,100 synonymous variants, 23 of which had a SpliceAI score above 0.5 suggesting an effect on splicing. A median of 229 variants were predicted to cause loss of function, including stop-gain, start-loss, frameshift and splice-acceptor or donor changes; 165 of those were classified as high confidence by LOFTEE. The frequency distributions for potentially deleterious variants were more similar among ancestry groups than those for common or less consequential variation.

Large structural variation was assessed more conservatively because short-read sequencing has limited ability to detect many copy-number changes, inversions, tandem repeats and mobile-element insertions. The researchers restricted their analysis to high-quality autosomal deletions and duplications longer than 10,000 base pairs, for which data were available for 587,475 participants. Individuals carried a median of 68 such variants, including 46 deletions and 22 duplications. About one in 159 people carried a deletion larger than 1 megabase and one in 55 carried a duplication larger than that. Although most copy-number variants were common, they were predicted to cause loss of function in about 25 genes per genome and complete copy gain in roughly five.

Common variation also left a substantial imprint on autozygosity. Adults in UK Biobank had an average of 3.65 unique runs of homozygosity at least 1.5 megabases long, compared with 4.00 in All of Us, but the figures varied markedly by ancestry group. All of Us participants classified as South Asian had an average of 5.27 such runs, while those classified as Middle Eastern had 7.14. Runs reached lengths of 242 megabases in UK Biobank and 248 megabases in All of Us, nearly encompassing an entire chromosome in some individuals. Based on a threshold of more than 90% chromosomal homozygosity, about one in 8,100 people was identified as a potential case of uniparental disomy or somatic loss of heterozygosity, although the investigators noted that the adult volunteer cohorts may underrepresent such conditions.

Analysis of 2,334 parent–offspring trios indicated that the average genome contained 61 constitutive de novo single-nucleotide variants or short indels, with individual estimates ranging from 30 to 113. The mean number increased by 1.17 per year of paternal age at the offspring’s birth and by 0.38 per year of maternal age, with confidence intervals of 1.07–1.27 and 0.27–0.50, respectively. Among variants that could be phased, approximately 75% to 80% were paternal in origin. The trios had relatively young parents, whose mean age at the offspring’s birth was 23 years in UK Biobank and 29 years in All of Us, so the estimate lies toward the lower end of the expected distribution. A further median of two probably mosaic mutations was detected; their number rose with the sampled individual’s age but not with parental age.

Disease-associated findings were more nuanced than a simple count of pathogenic variants might suggest. Considering ClinVar pathogenic or likely pathogenic variants in genes curated by the Gene Curation Coalition for monogenic disease, the average genome contained about one to three such variants when analysis was restricted to rare variants. Most were heterozygous in genes linked to recessive disorders, which is consistent with carrier status rather than predicted disease. People also carried about two to five rare high-confidence loss-of-function variants in these genes, though some may not produce true loss of function or may act through a different disease mechanism. Among variants associated exclusively with autosomal dominant disorders, around one in five to 11 people carried at least one rare ClinVar pathogenic or likely pathogenic variant.

ClinVar variants of uncertain significance were far more numerous, occurring roughly 30 to 90 times more often than pathogenic or likely pathogenic variants in the average genome. A person carried about 180 such variants in monogenic-disease genes, including approximately 50 with a minor allele frequency below 1%. The ancestry comparison was uneven: participants with European-like ancestry carried more pathogenic or likely pathogenic variants and thus were more likely to receive a molecular diagnosis, while participants in other groups carried more variants of uncertain significance. The authors linked this imbalance to differences in allele frequencies, limited reference diversity and unequal access to genomic testing, while cautioning that the analysis captured only variants already catalogued in ClinVar and therefore does not represent every uncertain variant in a genome.

Polygenic scores produced a similarly counterintuitive result. Among 28 scores calculated for common diseases in UK Biobank, the average participant ranked in the top 10% for three conditions and in the top 5% for 1.4 conditions. About one in five people with European-like ancestry had a score in the top 1% for at least one disease, compared with roughly one in three among the smaller non-European group. The latter result was driven mainly by a few scores with poorly aligned distributions across ancestries and was considered likely to reflect the limited performance of scores developed in one population when applied to another, rather than a simple difference in disease risk. The authors said that simultaneous screening with many polygenic scores could therefore increase overdiagnosis and burden healthcare systems, especially for conditions with low heritability.

These findings should be interpreted as estimates from UK and US adult volunteer cohorts using mostly blood-derived DNA, short-read sequencing and reference genome build GRCh38. The participants were not selected for a particular disease, but healthy-volunteer bias, age-related somatic changes and incomplete representation of global populations limit how closely the results define a universal “average” genome. African-like groups carried roughly one million more called small variants on average than European-like groups, potentially reflecting both additional variation and poorer alignment of reads from a reference genome derived predominantly from European-like individuals. Long-read sequencing, multiple tissue samples, complete telomere-to-telomere pangenome references and more geographically diverse cohorts will be needed to characterize variation that current methods miss. Even with those limits, the study offers a rare genome-wide baseline: every person carries extensive variation, but only a fraction has established clinical meaning, while the interpretation of rare and polygenic findings remains strongly dependent on ancestry, data quality and disease context.

Subject of Research: Individual-level variation in 879,461 whole-genome sequences from UK and US population biobanks

Article Title: The average individual genome: a survey of nearly 1 million whole-genome sequences

Article References: Wright, H. I. W., Beaumont, R. N., Fasham, J., Leslie, J. S., Chundru, V. K., Hall, T. S., Cannon, S. J., Pagnamenta, A. T., Baple, E. L., Wood, A. R., Murray, A., Jackson, L., Weedon, M. N., Patel, K., Hawkes, G., & Wright, C. F. (2026). The average individual genome: a survey of nearly 1 million whole-genome sequences. Genome Medicine, 18(1), Article 160. https://doi.org/10.1186/s13073-026-01792-x

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01792-x

Keywords: whole-genome sequencing, human genetic variation, population biobanks, rare variants, copy-number variants, de novo mutations, ClinVar, polygenic scores, genetic ancestry, genomic medicine, average, individual