Back

Proteomic and clinical impact of human knockouts in British South Asians

Carrasco-Zanini, J.; Pietzner, M.; Koprulu, M.; Zoodsma, M.; Williamson, A.; Hunt, K. A.; Manolias, A.; Walter, K.; Martin, H. C.; Finer, S.; van Heel, D. A.; Langenberg, C.

2025-10-15 genetic and genomic medicine
10.1101/2025.10.15.25337977 medRxiv
Show abstract

Human loss-of-function (LoF) variants affecting both copies of a gene ("human knockouts") provide a unique opportunity to directly study function and clinical impact of genes but are very rare in most populations sequenced to date. Here we study 1,569 British Bangladeshi and -Pakistani adults who were recalled for plasma sampling for proteomic profiling using three distinct technologies (covering >12,000 proteins) from 55k whole exome sequenced Genes & Health adults - a cohort enriched for rare, biallelic (homozygous) variants due to high autozygosity. We identified 199 individuals with rare homozygous predicted LoF genotypes (pLoF) for which the respective cis-protein was measured by at least one technology, and observed extreme (> 3SDs) cis-protein underexpression in 41 individuals (median z-score = -9.72 (range: -19.61 to -4.78) and overexpression in 2 individuals (median z-score = 8.1 (range 4.80 - 11.40)), representing 19% of these variants. For missense homozygotes, we observed 158 individuals with significantly under-expressed cis-protein (median z-score = -6.95 (range: -28.16 to -3.95)) and 62 individuals over-expressed, median z-score = 5.65 (range: 4.57 to 25.08)). The majority (62%) of LoF knockout genes with an identified cis-protein effect had evidence from 2 or more platforms, highlighting the high confidence nature of these discoveries. Systematic clinical assessment of human knockouts with strong evidence of an impact on cis-protein abundance through multi-source electronic health record linkage enabled identification of 1) knockout carriers with rare disease features based on phenotypic similarity, 2) novel rare disease-causing variants, 3) evidence for reclassification of genes and variants of uncertain significance from ClinVar and rare disease panels, and 4) novel gene-phenotype associations in humans. Based on high confidence examples, we developed a machine learning model that predicted 1 in 4 pLOF and 9 in 10 missense variants are likely benign. In summary, our study provides strong human-derived insights into the fundamental biology and clinical relevance of many genes and shows the value of proteogenomic studies of human knockout carriers.

Matching journals

The top 2 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.