Sample ForIntel Research Field Atlas — How the COVID-19 Vaccines-and-Variants Field Is Built

The atlas

A concentrated field with one dominant research cluster, a citation mass anchored on an early-2020 founding cohort, a compact UK-and-China-led institutional core, and a large protected-IP base held by academic and state research bodies — mature, consolidating, and still publishing.

This is a structural map of the COVID-19 vaccines-and-variants research field, across five field vectors. On concept structure, the field corpus — 595 distinct works — concentrates in a single core research cluster of 371 works, with a long tail of specialist clusters (clinical research, epidemiology, mental-health, long-term effects, vaccine coverage/hesitancy). On citation magnitude, the field's gravity wells are a tight early-2020 founding cohort — the clinical-characterization papers, the novel-coronavirus virology work, and the pivotal vaccine-efficacy reports — carrying tens of thousands of citations each. On institutional structure, a compact set led by Oxford, UCL, Hong Kong, Imperial, Harvard, Cambridge and the LSHTM produces a disproportionate share of the field's most-cited output. On time evolution, the field is a 2020 founding burst of 213 works that has consolidated year over year while continuing to publish. On IP translation, the field carries a large protected-IP base (~9,250 distinct patent families, 43% filed since 2022) whose holders are academic and state research institutions, heavily Chinese — China's PLA Academy of Military Medical Sciences leads (84), then the University of California, Tsinghua, and the French/Spanish/US state bodies (INSERM, CNRS, Pasteur, CSIC, NIH); the only company in the top 25 is Celltrion, and the commercial vaccine makers (Moderna, BioNTech) lead only the vaccine-specific cut. So who patents the field is a different set from who publishes it (Section 03) and from who productized the vaccines. The actionable structure is not whether the field is active — it plainly is — but where the centre of gravity and the specialist gaps sit, so a funder, department head or analyst can place reading, scouting and partnership attention deliberately. This Atlas reports field structure only and makes no scientific or clinical claim about COVID vaccines or variants.

• The field concentrates in one dominant research cluster, with a long specialist tail. The scholarly-publication corpus — 595 distinct field works — organizes into a single core research cluster of 371 works, with a long tail of adjacent specialist clusters: clinical-research studies (48), epidemiology (26), mental-health research (22), long-term-effects research (17), vaccine coverage / hesitancy, and computational drug discovery. The structural read: the field has a large, tightly-held centre of gravity with specialist work organized around it, not a set of evenly-sized sub-fields. The corpus pulls were capped, so every count is a floor — the read is the cluster structure, not an exact census.
• The citation mass concentrates on an early-2020 founding cohort. Ranked by citation magnitude (how heavily each work is cited), the field's foundational layer is a tight early-2020 cohort: the clinical-characterization papers (~31,000 and ~30,000 citations), the novel-coronavirus identification and cell-entry-mechanism virology work (~23,000 and ~21,000), and the pivotal vaccine-efficacy reports (the BNT162b2 and mRNA-1273 trials, ~15,500 and ~10,700). These are the field's gravity wells — where attention concentrates. Citation magnitude is a structural signal, not a statement that any work is scientifically correct.
• A small set of institutions produces a disproportionate share of the field. Aggregating the field corpus by producing institution (counts are floors on the field corpus), a compact leadership set dominates: University of Oxford (49 works), University College London (31), University of Hong Kong (29), University of Cambridge (22), Harvard University (22), Imperial College London (22) and the London School of Hygiene & Tropical Medicine (21), with the Chinese Academy of Sciences carrying an outsized citation footprint relative to its work count. The structural read: a UK-and-China-anchored institutional core produces the field's most-cited output — the natural places to scout talent or seed partnership.
• The field is a 2020 founding burst that consolidates — and is still publishing. By publication date the field's mass concentrates in a 2020 founding burst of 213 works that consolidates year over year (158 in 2021, 94 in 2022) — the canonical pandemic-research arc of an explosive founding cohort the field keeps building on. A separate current-activity confirmation shows the field is still actively publishing (200 works dated 2026 in the recency-sorted pull), so the consolidation is a maturation of the foundational cohort, not a cooling of the field. The recent biomedical-literature velocity corroborates continued output in both the vaccine and variant streams.
• The field translated into a large protected-IP base — and the IP is led by academic and state research bodies, not the vaccine companies. Reading the patent record at full strength — every distinct invention counted once worldwide — the field has translated into ~9,250 distinct patent families (43% filed since 2022). The striking structural finding: the IP is dominated by academic and state research institutions, heavily Chinese — China's PLA Academy of Military Medical Sciences leads (84 families), then the University of California (70), Tsinghua (63), the French state bodies INSERM (58) and CNRS, Fudan, the Chinese Academy of Sciences, Institut Pasteur, Spain's CSIC and the US NIH (all ~40–50). The only company in the top 25 is Celltrion — the commercial vaccine makers (Moderna, BioNTech) do not lead patent volume; they surface only in the vaccine-specific cut. So the field has three distinct actor sets: who publishes it (UK-led academia, Section 03), who patents it (China-and-state-led research bodies, here), and who productized the vaccines (the Western commercial makers). One framing caveat: this counts patenting activity (how many distinct inventions each holder files), not IP value.

In one line: the COVID-19 vaccines-and-variants field is a concentrated structure — one dominant research cluster of 371 works inside a 595-work field corpus, a citation mass anchored on an early-2020 founding cohort, a compact UK-and-China-led institutional core, a 2020 founding burst that has consolidated while the field keeps publishing, and a large protected-IP base (~9,250 patent families) whose holders are academic and state research bodies, China-led (PLA Academy, UC, Tsinghua, INSERM) — a different set again from who publishes or who productized the vaccines. The actionable structure is not "is the field active" — it is — but where the centre of gravity sits and which specialist clusters and institutions are the highest-leverage places to place reading, scouting or partnership attention.

How to read this atlas. This is a field-structure map — it reports how the COVID-19 vaccines-and-variants research field is organized (concept clusters, citation magnitude, institutional concentration, time evolution, and IP translation) and turns that into where to place field attention. One basis note so the sections reconcile: the cluster and institutional aggregates span the full retrieved record including current-year works, while the founding-era time curve (Section 04) is computed on the pre-recency dated subset — so the corpus totals and the year curve are deliberately on different bases. It is not a scientific or clinical assessment: it does not adjudicate whether any vaccine or variant claim is correct, rank therapies by efficacy, or make a medical recommendation — any scientific interpretation is the buyer's, with domain experts. A High confidence chip marks a directly observed structural signal; a Medium chip marks a signal that is real but rests on a capped pull, with the limitation stated inline. Every quantitative claim traces to a specific retrieved record; capped pulls are reported as floors, never rounded up. One structural layer — the directed citation graph (the who-cites-whom edges that would let us name bridge researchers) — is deferred as an operationally-heavy enhancement and named honestly in the closing section; citation magnitude is fully present, only the directed edges are deferred. The highest-stakes claim — the field's continued publication velocity — was independently re-verified against a second corpus.

01 · Concept-Cluster Structure

How is the field organized — one centre, or many? (Confidence: High.)

The first structural question is how the field divides into research clusters — because where the work concentrates tells a buyer where the field's centre of gravity is and where the specialist edges sit. Anchored on the field's core concepts and deduplicated to distinct works, the corpus is 595 distinct field works, and it does not divide evenly. One core research cluster of 371 works — the central SARS-CoV-2-and-COVID-19 research body — dominates, with a long tail of adjacent specialist clusters organized around it.

Figure — The scholarly-publication corpus (595 distinct works) concentrates in a single core research cluster of 371 works, with a long tail of specialist clusters. The pulls were capped, so every count is a floor; the read is the cluster structure, not an exact census.

Two structural reads matter. First, the field has a large, tightly-held centre — the core cluster is several times the size of any specialist cluster — so the typical work sits in the mainstream of the field rather than in a niche. Second, the specialist tail is where the field's distinct sub-conversations live: clinical-research methodology, epidemiological modelling, the mental-health and long-term-effects literatures, and the vaccine-coverage / hesitancy and computational-drug-discovery clusters. For a buyer deciding where to read or seed work, the centre is well-covered and crowded; the specialist clusters are the more legible places to find an under-attended edge. The whole map is built from capped pulls, so the cluster sizes are floors — the load-bearing read is the shape of the concentration (one dominant cluster, a long tail), which a deeper, uncapped corpus would sharpen but not overturn.

02 · Citation-Magnitude Ranking

Where does the field's citation mass concentrate? (Confidence: High.)

The second structural question is which works the field is built on — its foundational layer. We read this through citation magnitude: how heavily each work is cited, supplied inline with every record. The field's most-cited works are a tight, coherent early-2020 founding cohort.

Figure — Ranked by citation magnitude, the field's foundational layer is the early-2020 clinical-characterization and virology cohort, joined by the pivotal vaccine-efficacy reports. Citation magnitude is a field-structure signal — where attention concentrates — not a statement about the scientific merit or correctness of any individual work.

The structural read is the concentration: the field's citation mass is held by a small early-2020 cohort, an order of magnitude above the typical work, and that cohort spans both sub-areas — the clinical / virology foundation and the pivotal vaccine-efficacy reports. For a buyer, these are the works a newcomer to the field must read first and the lineages most downstream work descends from. The essential discipline: citation magnitude measures attention, not correctness — a heavily-cited work is a structural landmark, not an endorsed scientific conclusion, and this Atlas makes no claim about whether any of these works is right. One layer is deliberately deferred here: the directed who-cites-whom edges that would let us trace exactly which later works cite which foundational work — the citation lineage — are an operationally-heavy enhancement named in the closing section; the magnitude read above is complete.

03 · Institutional Concentration

Which institutions produce the field? (Confidence: High.)

The third structural question is who produces the field — because the institutional concentration tells a funder or a department head where the talent and the most-cited output sit. Aggregating the field corpus by producing institution, a compact leadership set carries a disproportionate share of the field's output.

Figure — A UK-anchored academic core leads on output (Oxford, UCL, Cambridge, Imperial, the LSHTM, King's), with Hong Kong and US institutions completing the leadership set. Work counts are floors on the field corpus. The Chinese Academy of Sciences carries an outsized citation footprint relative to its work count, driven by the foundational early-2020 virology cohort.

The structural read is a concentrated, UK-and-China-anchored institutional core producing the field's most-cited work. For a buyer, this is the map of where to scout talent, where the existing collaboration density is highest, and which institutions to approach for partnership. One honest boundary travels with this layer: the field corpus carried no usable funder / grant metadata — the funding-flow read that would name which funders underwrite the field is not recoverable from the returned records and is named as a boundary in the closing section, so the institutional read here is an output-and-citation concentration read, not a funding-flow read. And the work counts are floors (capped pulls), so the read is the relative concentration among institutions, not an exact ranking by one or two works.

04 · Time Evolution

How has the field grown — and is it still active? (Confidence: High.)

The fourth structural question is the field's trajectory over time — when it formed, how it has matured, and whether it is still active. By publication date the field has the canonical pandemic-research shape: an explosive 2020 founding burst that consolidates year over year.

Figure — The field's publication mass concentrates in a 2020 founding burst (213 works) that consolidates year over year (158 in 2021, 94 in 2022). The 2026 figure is on a different basis — a recency-sorted current-activity confirmation pull (200 works dated 2026) shown to confirm the field is still actively publishing, not to compare against the founding-era counts. Every count is a floor (capped pulls).

Two structural reads matter. First, the field is mature, not nascent: its foundational mass was laid down in 2020 and the field has been consolidating and building on it ever since — the year-over-year decline in new-founding-cohort volume is the normal maturation of a field whose canonical works are already written, not a sign the field has gone quiet. Second, the field is still actively publishing: the current-activity confirmation returns a full pull of works dated to the current year, and the recent biomedical-literature velocity (an independent second corpus) corroborates continued output in both the vaccine and the variant streams — the variant stream in particular shows rising recent biomedical activity. The honest framing: this is a maturing, consolidating field that remains live, and the buyer should read the consolidation as the field deepening around its founding cohort, not cooling. The highest-stakes claim here — that the field is still active — was independently re-verified against a second corpus, and it holds. One basis note so the year curve reconciles with the corpus total: the founding-era curve covers the dated, pre-recency cohort only — the current-year works from the recency pull and a residue of undated records sit outside it, so the per-year bars deliberately do not sum to the full 595-work field corpus.

05 · IP Translation

Has the field translated into protected IP — and who holds it? (Confidence: High.)

The fifth structural question is whether the research has translated into protected intellectual property, and which organizations hold it. Reading the patent record at full strength — full-text and classification, every distinct invention counted once worldwide — the field has translated into a large IP base: ~9,250 distinct COVID-19 patent families, 43% of them filed since 2022. And the answer to who holds it is the section's most counter-intuitive finding.

Figure — Distinct COVID-19 patent families per harmonized assignee (global, family-deduped). The IP is led by academic and state research institutions, heavily Chinese — China's PLA Academy, the University of California, Tsinghua, and the French/Spanish/US state research bodies (INSERM, CNRS, Pasteur, CSIC, NIH). The only company in the top 25 is Celltrion; the commercial vaccine makers (Moderna, BioNTech) lead only the vaccine-specific cut, not total patent volume.

Two structural reads matter. First, who patents the field is a different actor set from who publishes it — and from who sells the vaccines. Section 03's most-productive research institutions are UK-led (Oxford, UCL, Imperial); the patent leaders are China-and-state-led research bodies — the PLA Academy of Military Medical Sciences (84), the University of California (70), Tsinghua (63), Fudan, the Chinese Academy of Sciences, and the European state research bodies INSERM, CNRS, Institut Pasteur and CSIC, plus the US NIH. The famous commercial vaccine makers — Moderna, BioNTech — are not in the top 25 by total patent volume; they surface only when you narrow to vaccine-specific patents, where the University of California still leads and the company portfolios sit alongside state vaccine programs (the Gamaleya/Sputnik program, Sinovac, SK Bioscience, the US NIH). The structural read for a buyer: the field has three distinct maps — a research map (UK-led), a patent-IP map (China-and-state-led), and a vaccine-market map (Western commercial) — and conflating them is the classic mistake this layer prevents. Second, this read is gold-standard by construction: global, family-deduplicated (one invention filed across the US, Europe and the PCT counts once), classification-and-full-text scoped, and assignee-harmonized. The counts are exact distinct-family counts, not a floor. Two honest framings travel with it. First and most important: this measures patenting activity — how many distinct inventions each holder files — not IP value. Academic and state bodies file many narrow families across the whole research space; a commercial maker's few, foundational families (an mRNA-platform patent, say) can be worth far more than dozens of narrow ones. So read this as "who is most active in patenting the field," not "who owns the most commercially valuable IP" — a value-weighted overlay (forward citations, claim breadth) is the natural next layer. Second, a minor methodology residual: the full-text scope covers title + abstract (not the full claims, a cost choice), and a few large filers' sub-entities are hand-consolidated. Neither moves the structural picture — the academic/state-led pattern holds.

06 · What This Means For You

1 · Anchor any read of the field on the early-2020 founding cohort — then look to the specialist clusters for the under-attended edge. The citation mass is concentrated in a small, coherent founding cohort (the clinical-characterization, virology and pivotal vaccine-efficacy works), so a newcomer's reading list and a funder's baseline both start there. But the crowded centre is not where marginal attention pays off — the specialist clusters (long-term effects, vaccine coverage / hesitancy, computational drug discovery, the mental-health literature) are the more legible places to find an under-served edge.

2 · Scout talent and seed partnership at the compact UK-and-China-led institutional core. A small set of institutions — Oxford, UCL, Hong Kong, Imperial, Harvard, Cambridge, the LSHTM — produces the field's most-cited output, and the Chinese Academy of Sciences carries an outsized citation footprint relative to its volume. A funder or department head placing recruitment or collaboration attention should start with that core, and treat the high-citation-footprint-per-work institutions as the ones sitting closest to the field's founding layer.

3 · Read the field as maturing and live, not cooling — and target the still-rising sub-streams. The 2020 founding burst has consolidated, but the field is still publishing, and the recent biomedical velocity shows continued output in both sub-areas — the variant stream in particular shows rising recent activity. A buyer deciding where new money or attention earns the most should weight the still-rising sub-streams over the saturated founding-era topics.

4 · Treat the three maps as three different things — research, patent IP, and vaccine market do not overlap the way buyers assume. Who publishes the field is UK-led academia (Section 03); who holds the patent IP is China-and-state-led research bodies (the PLA Academy, the University of California, Tsinghua, INSERM/CNRS/Pasteur/CSIC, the NIH); who holds the vaccine market is the Western commercial makers (Moderna, BioNTech) who don't even crack the top 25 by total patent volume. For licensing, partnership, competitive or freedom-to-operate scouting, use the patent-IP map — not the publication map and not the brand-name vaccine list. The single most consequential read: a state research apparatus (China's, plus the European state bodies) owns more of the field's protected IP than the household-name vaccine companies do.

5 · Commission the directed citation-graph layer to convert this structural map into a connectivity map. This Atlas maps where the clusters, the citation mass and the institutions sit; the natural next step is the directed who-cites-whom layer — which would identify the bridge researchers who connect the vaccine and variant sub-areas and the citation-co-occurrence linkage between clusters. That layer is named as an operationally-heavy enhancement in the closing section (the magnitude read is complete; only the directed edges are deferred). Commission it to turn the map of where things sit into a map of how they connect.

Scope, Confidence & What a Deeper Engagement Adds

This Research Field Atlas maps five field vectors — concept-cluster structure, citation-magnitude ranking, institutional concentration, time evolution, and IP translation — the first four directly observed against the returned scholarly-publication record (with the highest-stakes claim, the field's continued publication velocity, independently re-verified against a second corpus), and the fifth read from the patent record at gold standard. The boundaries below are named with the specific reason and the work that closes each. They are diligence boundaries, not findings, and are never presented as such. Above all: this Atlas reports field structure, not scientific merit — it does not adjudicate whether any vaccine or variant claim is correct, rank therapies by efficacy, or make a medical recommendation.

• The directed citation graph is a deferred, operationally-heavy enhancement. This Atlas reads citation magnitude (how heavily each work is cited) fully and inline. The directed who-cites-whom layer — the citation-co-occurrence linkage that shows which bodies of work cite together, the bridge-researcher identification (researchers cited across both the vaccine and variant sub-areas), and the reference-lineage walk — requires fanning a per-work citing-set traversal out across every foundational work, and that fan-out exceeded the analysis window across repeated reproductions. This is a near-term capability enhancement (a lighter or asynchronous traversal), not a data absence: the citation counts the structural read rests on are present, and the magnitude, cluster, institution and evolution reads are complete. We name no bridge researchers and draw no who-cites-whom edges, because that specific data was not captured. Closing it: a lighter-fan-out or asynchronous citation-graph traversal to add the directed edges and the bridge-researcher read.
• The patent IP-translation read is gold-standard — two small residuals named for completeness. The IP-translation layer (Section 05) reads the patent record at full strength: global, family-deduplicated, classification-and-full-text scoped, and assignee-harmonized — so the holder counts are exact distinct-family figures, not the title-only floor a quicker read would give (an earlier US-title-only pass mis-ranked this layer as commercial-vaccine-maker-led; the gold-standard read corrects it to academic/state-led). Two honest residuals, neither moving the structural picture: (1) the full-text scope covers title + abstract, not the full claims/description (a deliberate cost choice); a patent that mentions COVID only deep in its claims is not counted. (2) Assignee harmonization is strong but imperfect — a few large filers' sub-entities (and one state vaccine program fragmented across its named inventors) are consolidated by hand in the read. Closing them: a claims-level full-text pass and entity-resolution layer — a marginal refinement, not a gap.
• The corpus pulls were capped — every publication count is a floor. The concept, magnitude and institution pulls each returned the maximum number of items allowed, so the field-corpus size (595 distinct works), the cluster sizes, the institutional work counts and the per-year volumes are floors, not a complete census. The structural reads — one dominant cluster, an early-2020 citation mass, a compact institutional core, a founding-burst-then-consolidation arc — are unaffected and are the load-bearing reads; the exact volumes are bounded. Closing it: an uncapped corpus pull to quantify the full field and sharpen the exact counts.
• The funder / grant metadata was absent — the funding-flow read is not recoverable here. The returned field corpus carried no usable funder / grant metadata, so the institutional read is an output-and-citation-concentration read, not a funding-flow read — this Atlas does not name which funders underwrite the field. Closing it: a funder-metadata-bearing corpus pull to add the funding-flow layer.
• The citation read runs on a single scholarly corpus by construction. No independent second corpus indexes the citation graph at this grain, so the citation-magnitude and (when added) directed-graph reads are single-corpus by construction. The publication-velocity claim was cross-verified against an independent biomedical corpus; the citation-structure claims could not be. Closing it: a second citation-graph index to corroborate the citation-structure reads.

This is a field-structure read at the Research Field Atlas tier, built on the scholarly-publication record (concept clusters, citation magnitude, institutional structure, time evolution), a biomedical-literature cross-corpus, and the patent record (the IP-translation layer). The natural next step is a deeper engagement that turns this map into a connectivity map: (1) the directed citation-graph layer to identify the bridge researchers and the cluster linkage; (2) a patent-value overlay on the IP-translation read (forward-citation weighting, claim breadth, family size) to turn the filing-volume ranking into a value-weighted one; and (3) an uncapped, funder-metadata-bearing corpus to quantify the field and add the funding-flow layer. To commission it, reach the ForIntel desk directly at forintel@foragentis.com.