Genetics papers went from one institution to nine

This site rests on 671 distinct papers. Every published variant and every condition page cites at least one, and those citations are the whole basis on which anything here claims to be true.

So there was an obvious question to ask of them: which universities are doing this work, and what is each of them working on?

The answer we got back was not the one we went looking for.

What we did

Papers carry author affiliations, and Europe PMC returns them. 658 of our 671 sources have at least one, 13,197 affiliation strings in all, and after matching, 599 papers yield at least one identifiable institution.

Two rules did most of the work, and both are worth stating because they decide what the numbers mean:

• An institution is counted once per paper, never once per author. A 107-author paper would otherwise make one department look like forty.
• Teaching hospitals are not folded into their universities. Massachusetts General Hospital and Harvard Medical School stay separate. Merging them is defensible, and it is a judgement about institutional structure that differs by country, so we left it to the reader.

The first result was that everybody looked identical

Oxford, Harvard, Karolinska, Groningen, Helsinki, Copenhagen — the trait lists came back nearly the same. Atrial fibrillation. Age at menarche. Colorectal cancer. Daytime napping.

That is not what six different institutions research. It is one artefact repeated six times, and finding out why took the survey somewhere better than it was going.

The reason

A consortium paper contributing 43 variants hands the same 43 traits to every institution on its author list. If that list has sixty institutions on it, sixty institutions now appear to study exactly the same thing.

So we measured how common that is. Median number of institutions per paper, by decade of publication:

The median genetics paper in our sources was a single-institution paper for two decades. In the 2020s it is a nine-institution paper. The widest one has 378 institutions on it — a saturated map of the common variants associated with human height — and it contributes zero variants to this site, because it maps a trait rather than reporting positions we index.

How much of the catalogue this affects

Sixty-nine papers — one in ten — carry nearly a quarter of everything the catalogue holds. And each of those sixty-nine spreads its subject matter across every institution that signed it.

A ranking here would be a ranking of consortium membership

This is the part that would have been genuinely misleading if we had published the first table and stopped.

Below is each institution's paper count, and beside it the share of those papers that are narrow — five institutions or fewer, where authorship still means a research group rather than a signature on a meta-analysis.

Ninety-three percent of the Broad Institute's appearances in our sources are on wide papers. For Karolinska and Copenhagen it is ninety-four. A league table built on this data would mostly be measuring which institutions join the most consortia — which is a real and interesting thing about modern science, and it is not what anybody would think they were reading.

What you can still say

Restrict to the narrow papers and the profiles separate immediately:

• deCODE Genetics (Iceland) — prostate cancer, thyroid cancer, PSA levels
• Institute of Cancer Research (UK) — prostate, colorectal and breast cancer
• Erasmus Medical Center (Netherlands) — the eye: optic disc parameters, vertical cup-disc ratio
• University of Tokyo — haematological and biochemical traits, urate
• Queensland Institute of Medical Research (Australia) — optic nerve measurements, endometriosis
• Wellcome Trust Centre for Human Genetics (Oxford) — Crohn's disease, type 2 diabetes, adiposity

Those read like places with a subject. They are also, in every case, a much smaller number of papers than the headline count — deCODE has 15 narrow papers, the Institute of Cancer Research 11.

What this does not mean

It does not mean an institution absent from this list is absent from genetics.

The concrete example: the University of Texas at Austin appears in zero of our 671 sources. Texas appears plenty — UTHealth Houston on 12 papers, Baylor College of Medicine on 10, MD Anderson on 9 — but all of it is medical schools and cancer centres.

That is a fact about us. Our collector reads the GWAS Catalog, which indexes human disease-and-trait association studies. Evolutionary genetics, population genetics and computational biology do not produce the kind of paper we index, so a department strong in those is invisible here no matter how good it is. The Texas pattern — every hit a hospital — is that bias showing its shape.

So this survey measures appearance in one catalogue's source list. It is not a measure of research quality, output or importance, and any reading of it as a ranking of universities is a misreading.

One more thing, because it nearly went unnoticed

The first version of this analysis matched institution names with a pattern that required a word boundary immediately after universit. There is never a word boundary there — "University" carries on — so the pattern could not match the word it was written for. It silently dropped 262 of 671 papers, including every paper affiliated only to the University of Cambridge.

Nothing failed. No error was raised. The output looked entirely reasonable, and the top of the table was a list of hospitals. It was only obviously wrong because Cambridge appeared 288 times in the raw text and nowhere in the results.

We mention it because the same shape of error is what a reader should assume is possible in any counting exercise, including this one, and because a survey that reports its own near-miss is easier to trust than one that does not.

Why any of this matters when you read a paper

If you look up the study behind a variant on this site and find it has 171 authors from 100 institutions, that is not padding. It is what statistical power costs now. Finding a variant that shifts risk by three percent requires more people than any one university can recruit, so the recruiting is pooled and the authorship follows.

The trade is that a paper stops being the product of a place. Which is why we can tell you what the 671 papers behind this site found, and cannot honestly tell you what any one university is working on.

Every figure here was measured against this site's published sources on 8 September 2026 and moves as the catalogue grows. See where our data comes from for what those sources are.