A study led by researchers at the Icahn School of Medicine at Mount Sinai finds that single-cell atlases do not fairly represent the world’ populations. Published in Cell Genomics, the analysis found that people of European ancestry were consistently overrepresented, Asian and Latino individuals were underrepresented, and a large portion of samples had no recorded ancestry at all.
The team examined more than 13,500 samples from three major resources: the Human Cell Atlas, the Human Tumor Atlas Network, and the PsychAD Consortium. They compiled reported ancestry, race and ethnicity, and sex data, then compared each dataset against global population figures, U.S. cancer incidence data, and disease-specific references.
The gaps appeared across all three resources. Nearly 70 percent of Human Cell Atlas samples lacked ancestry information. Among samples that did report it, European ancestry individuals were overrepresented roughly sixfold relative to global expectations, while Asian, African, and Latino individuals were underrepresented. Tumor samples in the Human Tumor Atlas Network were about 69 percent European, and the PsychAD brain dataset was nearly two-thirds European. Several cancer types also showed sex-based differences beyond what disease incidence would predict. The European overrepresentation in the Human Cell Atlas persisted even under the most conservative assumptions about missing data.
Senior corresponding author Kuan-lin Huang said the team wanted to determine “do these maps represent people from different populations fairly, or are some groups being left out?” He added that “some of the most important datasets are not as representative as they should be,” and that missing ancestry information “can be passed into AI models trained on the datasets, often without users of the AI models realizing it.”
The authors note the study relied on demographic data reported in public datasets rather than direct measurement of genetic ancestry, and that broad categories cannot capture the full complexity of identity. The findings may not extend to every single-cell dataset in the field. The team has also produced a checklist covering recruitment planning, demographic recording, sample balance, and evaluating whether AI models perform equally across ancestry and sex groups, and plans to expand this type of audit to additional datasets over time.
The team examined more than 13,500 samples from three major resources: the Human Cell Atlas, the Human Tumor Atlas Network, and the PsychAD Consortium. They compiled reported ancestry, race and ethnicity, and sex data, then compared each dataset against global population figures, U.S. cancer incidence data, and disease-specific references.
The gaps appeared across all three resources. Nearly 70 percent of Human Cell Atlas samples lacked ancestry information. Among samples that did report it, European ancestry individuals were overrepresented roughly sixfold relative to global expectations, while Asian, African, and Latino individuals were underrepresented. Tumor samples in the Human Tumor Atlas Network were about 69 percent European, and the PsychAD brain dataset was nearly two-thirds European. Several cancer types also showed sex-based differences beyond what disease incidence would predict. The European overrepresentation in the Human Cell Atlas persisted even under the most conservative assumptions about missing data.
Senior corresponding author Kuan-lin Huang said the team wanted to determine “do these maps represent people from different populations fairly, or are some groups being left out?” He added that “some of the most important datasets are not as representative as they should be,” and that missing ancestry information “can be passed into AI models trained on the datasets, often without users of the AI models realizing it.”
The authors note the study relied on demographic data reported in public datasets rather than direct measurement of genetic ancestry, and that broad categories cannot capture the full complexity of identity. The findings may not extend to every single-cell dataset in the field. The team has also produced a checklist covering recruitment planning, demographic recording, sample balance, and evaluating whether AI models perform equally across ancestry and sex groups, and plans to expand this type of audit to additional datasets over time.