What is counted
For each NIH funding component (24 institutes and centers and the Office of the Director), ScienceDex counts the component's FY2026 projects in NIH RePORTER, one per core project number, with the component's own dollars. A project funded by several components counts on each of their pages and once in the NIH total.
A shared dataset is linked to a project when its repository record names the project's grant number, or when it comes from a paper that NIH RePORTER links to the grant in any year. The institute page counts datasets, the projects with at least one, and the share of the institute's FY2026 funding those projects hold.
Where the links come from
Every source was searched once for all 25 components, so each institute is counted from the same records.
| Source | What links a dataset to a grant |
|---|---|
| DataCite | Records whose funding field names the grant number, found by an exact award search and by the institute's name. |
| PubMed (MEDLINE) | Data accession numbers listed on a paper NIH RePORTER links to the grant. |
| GEO | Series and curated datasets whose GEO record cites a linked paper. |
| DataCite, other spellings and papers | Records that name the grant in another spelling, or that name a linked paper. |
| Protein Data Bank | Entries whose funding field names the grant, and entries whose primary citation is a linked paper. |
| BioProject, dbGaP, DANDI, OpenNeuro, Zenodo, figshare | Records whose funding or grant field names the grant. |
| ShareScore | Dataset records whose ShareScore entry names the grant. |
In free text and in other funders' fields, a five-digit serial counts only with an activity code before it (R01, P30, P2C), because a bare five-digit number can be a ZIP code. A record that names more than 20 linked papers, such as a cohort study that lists the papers written with its data, does not link through them.
What one dataset is
A dataset is one deposit, so each of these counts as one dataset: the versions of a record, a GEO series with its own BioProject, a GEO SuperSeries with its SubSeries, a Zenodo record with its concept DOI, and the per-sample files of one TOPMed study. GEO series filed under a shared umbrella BioProject, as ENCODE files them, stay separate deposits.
On an institute page, “Where they are deposited” counts each dataset once, under the repository that holds it. A dataset found two ways counts once in the total and under both ways in “How they were found”.
What is set aside
The count keeps a link only when it runs through the project's own award. Two kinds of links credit many projects at once, so they are set aside and listed in the downloads with a flag.
- Center, resource and training awards. Every investigator such an award serves may cite it. The codes: P30, T32, T35, K12 and R25; UL1, KL2 and TL1 (CTSA hubs); P51, P40 and U42 (primate, animal and stock centers); P41 (technology resource centers); S10, G20 and C06 (shared instruments and facilities); and T15, T34, T37, T90, R90, TL4, D43 and D71 (other institutional training awards). Research mechanisms such as P01, P20, P50, U19 and U54 count as the project's own award.
- Consortium papers. A link made only through a paper that more than 1,000 records from the first three sources cite. One paper is cited that widely: the ENCODE Consortium's 2012 paper (PMID 22955616).
Across NIH, 66,951 datasets remain after both are set aside.
Kept outside the totals
These records name an institute or one of its grants but are not counted. The links file lists them, with the code below in its group column and the reason in why_not_counted.
- Datasets that name only grants with no FY2026 record (
older_institute_grant). - Datasets that name an institute but none of its grants (
institute_named_no_grant). - Journal supplement files on a publisher's figshare portal (
journal_supplement). - Data management and sharing plans that a repository typed as datasets (
sharing_plan).
The NCI and NIBIB totals
The NCI and NIBIB pages also open with an “Award-linked datasets” total. Both are built from this data version, one row per project and deposit, so they count the same deposits as this page. A page set to the awards an institute administers, rather than the projects it funds, counts that list of projects, so its total can differ.
Downloads
Each institute page offers two CSV files in the section on datasets linked to its grants. The datasets file has one row for each dataset that section counts, so its number of rows is the section's total, and its repository column gives the section's “Where they are deposited” counts. When a page counts the awards an institute administers, the total at the top of the page can differ, and the section's note gives both. Columns: deposit_group (the dataset), the repository it is counted under, title, dataset URL, record_ids (its records, such as a GEO series and its BioProject), the FY2026 grants it is linked to, administered_by (the institute that administers each of those grants, in the same order), how_linked and PMIDs. A cell that lists several values separates them with semicolons.
The links file has a row for every link found, counted or not. Each row is one dataset record and the grant that links it, so a dataset with several records or grants has several rows. Columns: counted (yes or no), why_not_counted, group, grant, administered_by, recipient organization, the institute's FY2026 dollars, the RePORTER record, dataset_id, deposit_group, repository, title, dataset URL, how_linked, in_first_search (yes for links from the first three sources above), the grant text as the record gives it, PMIDs, and the p30_center_grant, institutional_award and consortium_paper_only flags. To get the section's total from it, keep the rows where counted is yes and count each deposit_group once.
An institute's files cover every FY2026 project it funds, including projects another institute administers; administered_by names that institute. Each grant page offers that grant's rows of the links file. When a cell's text would begin with =, +, - or @, the file puts an apostrophe in front of it, so spreadsheets show the text instead of running it as a formula.
Versions and checks
Counting rule sciencedex-grant-linked-datasets-1.0.0. The links come from ScienceDex's grant-to-dataset report for all 25 components. Data version nih-grant-dataset-links-fy2026-1.0.0 is built from report version 1.0.0. A published version is never changed; new sources or rules mean a new version. Before a version is published, every file is checked against the hashes the report records, the stored rows must rebuild each file byte for byte, and every institute's counts and the NIH-wide totals must equal the report's.