Data and Methodology

This page documents how the ATMP Research Platform is built: the data sources, how publications are identified as ATMP-related, how technology sub-fields are defined, how metrics are computed, and what the known limitations are. Every number on the platform should be interpretable using the information here.

What are ATMPs?

Advanced Therapy Medicinal Products (ATMPs) are a class of medicines for human use defined under EU Regulation 1394/2007 and assessed centrally by the European Medicines Agency (EMA) through its Committee for Advanced Therapies (CAT). There are four statutory categories:

This platform covers research related to all four ATMP categories. The scope extends beyond approved products to include the underlying enabling technologies — CRISPR genome editing, iPSC reprogramming, lentiviral vectors, and similar foundational science. Publications are identified by their MeSH term assignments, not by regulatory approval status.

Publication data

Data source

Publications are retrieved from Dimensions (Digital Science) using the Dimensions Search Language (DSL) API. Dimensions aggregates records from PubMed, Crossref, and other major bibliographic databases, enriched with MeSH assignments, institutional affiliation data, citation counts, funder names, and linked clinical trial records. All publication data was downloaded in May 2026.

Identification strategy

The central challenge in a bibliometric study of ATMPs is defining a principled research boundary. Free-text search is inadequate: the same concept appears under dozens of linguistic variants across journals, years, and languages. This platform uses MeSH (Medical Subject Headings) term classification instead, for three reasons:

1. MeSH terms are assigned by trained human indexers at the US National Library of Medicine, providing a consistent controlled vocabulary across all PubMed-indexed literature.
2. A publication about "CAR T-cell therapy" and one about "chimeric antigen receptor lymphocytes" will both carry the same MeSH descriptor regardless of how authors phrased their title.
3. Dimensions exposes MeSH assignments for all PubMed-indexed publications and supports filtering by canonical descriptor name.

A publication is classified as ATMP-related if Dimensions has assigned to it at least one of the 112 classified ATMP MeSH descriptors described below.

:::note[Query design] Only canonical DescriptorName values are used in API queries, never MeSH entry terms (synonyms). Dimensions indexes publications by canonical descriptor; querying synonyms is redundant and risks false positives. Each of the 112 descriptors is queried independently and results are deduplicated by Dimensions publication ID. :::

Country and institutional attribution

Country and institutional affiliation are taken from Dimensions' parsed affiliation records. Each publication can carry multiple country attributions (one per contributing institution). The unit of analysis in all cross-country comparisons is the paper-country pair, not the unique publication:

• A paper co-authored by Swedish and German researchers is counted once for Sweden and once for Germany.
• Author leadership (first-author country) is computed from per-paper affiliation arrays, where author order follows the Dimensions JSON array order reflecting the published byline.
• The global baseline in all Sweden-versus-global comparisons excludes Swedish papers to avoid contaminating the reference distribution with the quantity being compared.

Known limitations

• Affiliation parsing is imperfect. Some affiliations are unresolved or incorrectly attributed, particularly for institutions in countries with non-standard address formats.
• Conference papers and preprints are included if they carry at least one ATMP MeSH term in Dimensions.
• MeSH indexing typically lags six to eighteen months behind publication date. Publications from 2024 onward may be systematically undercounted relative to earlier years.
• MeSH terms are applied only to PubMed-indexed publications. Publications not in PubMed (some engineering and materials science journals, for example) will not appear in this corpus even if scientifically relevant.

ATMP research classification

Vocabulary source

The ATMP classification is built on the MeSH 2026 descriptor vocabulary released by the US National Library of Medicine (31,110 terms), parsed from the NLM XML distribution. Each descriptor record includes a canonical name, hierarchical tree numbers, a scope note (definition), and a list of entry terms (synonyms).

Classification approach

MeSH terms were selected in two passes, followed by explicit exclusion review.

Pass 1: Keyword matching. Each descriptor's canonical name and entry terms were searched against a curated set of ATMP-relevant keywords spanning gene therapy, viral vectors, genome editing, CAR and other immune cell therapies, stem cells, tissue engineering, and manufacturing technologies. The first matching keyword category is recorded.

Pass 2: Tree-number matching. Two MeSH sub-trees were identified as wholly ATMP-relevant and all their member descriptors included:

• E02.095.260 (Gene Therapy, within the MeSH analytical/therapeutic techniques hierarchy)
• E02.095.465.430 (Immunotherapy, Adoptive)

Explicit exclusions. Two descriptors were removed despite triggering keyword matches, because domain review confirmed they are unrelated to ATMPs:

• Cell-Free System (matched "stem cell" in extended entry terms; describes in vitro cell-free translation, not therapy)
• Nevus, Sebaceous of Jadassohn (matched "organoid" in entry terms from clinical literature; describes a congenital skin hamartoma)

Certainty tiers

Both tiers are included in all platform analyses.

What is intentionally excluded

The following categories are not classified as ATMP-related, even where superficial MeSH overlap exists:

:::tip[Note on mRNA vaccines] The descriptor mRNA Vaccines is included because mRNA-based delivery is central to several ATMP development programmes, even though approved COVID-19 mRNA vaccines are not ATMPs under EU Regulation 1394/2007. Users who wish to exclude general vaccine research should apply additional domain filters. :::

Technology domain classification

Overview

To enable sub-field comparisons, each of the 112 ATMP MeSH terms is assigned to one of eight technology domains organised in two conceptual categories.

Fundamental Technologies cover the core biological mechanisms underlying ATMPs: how genetic material is edited and corrected, how cells are reprogrammed toward therapeutic identities, how therapeutic cargo is delivered into cells and tissues, and how engineered cells are designed to sense and respond to their environment.

Enabling Technologies cover the tools and platforms that support ATMP development without being the therapeutic mechanism themselves: how cells are characterised and quality-controlled, how they are expanded and processed outside the body, how they are tested in model systems before clinical use, and how they are fabricated into final products.

Domains at a glance

Assignment rules

Each MeSH term receives a primary domain assignment. A small number of terms with clear dual relevance also carry a secondary domain assignment (29 terms in total). In platform analyses, a paper's domain membership is determined by the primary assignments of its indexed MeSH terms. A paper indexed under terms spanning multiple domains appears in each of those domains.

Full classification table

The table below lists all 112 MeSH terms included in the ATMP research classification, together with their technology category, domain, and research sub-area.

Citation metrics

Raw citation counts

Citation counts are sourced from Dimensions and represent the total number of forward citations each publication has received from other works in the Dimensions database as of May 2026. Self-citations are not excluded.

Relative Citation Ratio (RCR)

The Relative Citation Ratio (Hutchins et al., 2016, PLOS Biology) is a field- and time-normalised citation metric from the NIH iCite database. An RCR of 1.0 means a paper has been cited at the same rate as the average paper in the same field and year; values above 1.0 indicate above-average citation impact within the field.

iCite computes RCR by comparing each paper's citation rate against the co-citation network of papers it most closely resembles, adjusting for both research field and publication year. This makes RCR more informative than raw citation counts for cross-field comparisons, where citation norms vary substantially.

:::note[RCR coverage] Approximately 93% of ATMP publications through 2022 have an RCR value. Publications from 2023 onward are largely excluded because iCite requires at least two years of citation accumulation to compute a stable rate. iCite data was joined to the ATMP corpus via DOI. Papers with a missing or zero RCR are excluded from all RCR-based analyses. :::

Science and commercial potential scores

Two scores derived from citation network analysis estimate the future impact trajectory of publications. Both come from the dataset published by Masclans et al. (2025), which provides pre-computed scores for tens of millions of publications.

:::note[Interpretation] Both scores are probabilistic estimates of future behaviour based on structural position in citation networks at the time of scoring. They reflect relative indicators, not precise predictions. Papers without a score are excluded from the respective analysis. :::

Altmetric coverage

Source

Altmetric data is retrieved from the Altmetric Details API (Altmetric.com / Digital Science) for all ATMP publications with a DOI. Coverage depends on Altmetric having indexed the publication. Data was downloaded in May 2026. For details on the data schema, see the Altmetric Explorer API documentation.

Coverage types

Patent jurisdiction classification

Patent citations are grouped by filing office to allow regional comparison.

:::note[Sweden is additive] Swedish patents (PRV filings) are counted in both the "Rest of Europe" group and a dedicated Sweden column. This is intentional: it allows Sweden to be compared against its regional peer group without needing to subtract Swedish contributions from the group total. :::

Policy and guideline source classification

242 unique policy document sources and 997 unique guideline sources were manually classified by geographic scope using known institutional identity and web search. Each source was assigned to one of the following categories, with a certainty level (high or medium):

Clinical trials

Source

Clinical trial metadata is retrieved from Dimensions via the clinical trials endpoint, drawing on ClinicalTrials.gov and WHO ICTRP registrations. Dimensions links publications to registered trials via citation and metadata matching.

Dimensions identifies links through two mechanisms: explicit citation of a trial registration number in the publication text, and algorithmic matching on shared identifiers and metadata. A single trial can be linked to multiple publications, and a single publication can be linked to multiple trials.

Phase classification

Translation rate

The translation rate for a publication group is defined as: the number of publications linked to at least one registered clinical trial, divided by the total number of publications in the group. Trial geography is based on the country of the registering organisation; a single trial can be attributed to multiple countries.

Funder information

Source

Funder names are extracted from Dimensions publication metadata. Each publication can carry multiple funder records. The ATMP corpus contains 6,092 unique funder name strings.

Classification

:::note[Coverage limitation] Of the 6,092 unique funder strings, 252 have been explicitly classified. The long tail of funders with 15 or fewer papers is intentionally left as unknown. Classification is concentrated in high-volume funders that account for the majority of funded publications. Dimensions funder strings often differ from assumed short forms (for example, "Wellcome Trust Ltd" rather than "Wellcome Trust"), meaning some known funders may be missed due to name variation. :::

Sample restrictions

The following exclusions apply consistently across all platform analyses:

• Publications with a missing year are excluded from time-series analysis.
• Publications with a missing or zero compot score are excluded from commercial potential analysis.
• Publications with a missing or zero RCR value are excluded from scientific quality analysis.
• Publications without a scipot match are excluded from science potential analysis.
• Records with a missing country code are excluded from country-level comparisons.
• The global baseline in all Sweden-versus-global comparisons excludes Swedish papers to avoid contaminating the reference distribution.

Platform

This platform is built with Observable Framework and Apache DuckDB. Charts use Observable Plot. Data queries run in a DuckDB instance (in-browser WASM for local use; remote server for the live deployment).

How to cite