OpenAlex research evolution, 2021–2025

The map after the shock.

Five years ago the center of gravity sat near pandemics, inference, and emergency relevance. By 2025 the same map leans toward equations, AI, batteries, frameworks, and machinery.

The surprise is not just that fields change. It is that they change in different ways: some walk across the map, some flee a single cluster, and some barely move at all while their internal mix quietly flips.

The full UMAP, filtered year by year. The colors are primary fields. The shape of the map stays stable; the density and emphasis do not.
Papers analyzed
15,068
Years included
2021–2025
Biggest move
Mathematics
Most visible post-COVID turn
Medicine
What the axes mean

This map is not random. It has a north, a south, an east, and a west.

The vertical dimension runs from living systems at the top—cancer, metabolism, genomics, single-cell biology—to formal systems at the bottom—equations, optimization, numerical methods, and physics-like structure. The horizontal dimension runs from AI-and-vision-heavy work on the left to materials, batteries, catalysts, and device engineering on the right.

That matters because a field’s drift becomes legible in plain English. A move down means a field is becoming more mathematical or systems-oriented. A move right means it is becoming more device-, energy-, or materials-oriented. A move up-left means it is leaning back into biology or perception.

These are inferred semantic poles from the map extremes, not official UMAP axis labels. Read them as two continuous dimensions: left ↔ right and top ↔ bottom.
Who moved farthest

The largest relocation belongs to mathematics.

Mathematics6.56Chemistry1.69Medicine1.62Biochemistry, Genetics and Molecular Biology1.29Engineering0.80
Trajectory view

Most fields do not wander. They commit.

When you track the citation-weighted centroids year by year, the big movers do not jitter. They travel in a direction with intent. Mathematics plunges. Medicine drops out of the pandemic band. Chemistry slides into frameworks and sensors. Engineering nudges rightward—less dramatic in geometry, but meaningful in composition.

Perception-to-Production axisLife-to-Logic axisAI, vision, perceptionmaterials, energy, devicesliving systems, biologyequations, formal systems2021Mathematics2021Chemistry2021Medicine2021Engineering
Hover, focus, or tap any point for year-level context. The horizontal line runs from perception-heavy work to production-heavy work; the vertical line runs from life-science territory down toward logic, equations, and formal systems.
Perception-to-Production axis Left means the field is sitting closer to AI, vision, and perception work. Right means it is drifting toward materials, energy, catalysts, batteries, and devices.
Life-to-Logic axis Up means the field is closer to living systems like cancer, genomics, and metabolism. Down means it is moving toward equations, optimization, physics-like structure, and formal methods.
Act I — The longest walk

Mathematics leaves the pandemic corridor and heads for the machine room.

In 2021, the most-cited mathematics papers in this corpus sit surprisingly high on the map, close to medicine and biostatistics. The neighborhood is full of statistical inference, data, and COVID-adjacent vocabulary. The dominant cluster owns 48.1% of the field. By 2025 that share collapses to 5.0%, while the equations-heavy cluster rises to 59.1%.

That is not a small edit. It is a relocation. The field’s weighted centroid moves 6.56 map units down-right, the largest shift in the corpus. The winning vocabulary changes too: equations, differential, fractional, numerical, solutions rise; covid-, epidemiological, pandemic, sars-cov-, impacts fade.

For a lay reader, the story is simple: the top-cited mathematics papers stop standing next to hospitals and policy dashboards and start standing next to control systems, equations, and optimization. Mathematics moves from measuring the emergency to rebuilding the engine.

Representative titles 2021: Statistical Issues in Drug Development
2025: Dual approach artificial neural networks-fractional order operator to enhancing in vitro fertilization predictions and success measure
The mathematics cohort, filtered year by year and colored by cluster. Watch the center of gravity slide out of the pandemic-statistics cluster and settle into the equations-and-systems neighborhood.
20212025Mathematics trajectory

The long diagonal fall is visible even in miniature: 2021 sits near pandemic inference, 2025 nearer equations and systems.

Medicine, year by year. The pandemic cluster still exists in 2025, but it no longer dominates the field’s center of gravity.
20212025Medicine trajectory

The motion is smaller than mathematics, but the drop out of the pandemic band is unmistakable.

Act II — The field everyone expected to move

Medicine does not leave biology. It leaves emergency mode.

In 2021, medicine is anchored by the pandemic cluster. COVID-shaped language dominates the field’s leading edge: covid-, sars-cov-, epidemiological, infections, virus. That cluster accounts for 56.3% of the field. By 2025 it falls to 14.9%, while the data-rich biology cluster rises to 45.0% and the AI-heavy cluster rises to 18.7%.

The new vocabulary is revealing. The gains are artificial, intelligence, education, healthcare, cancer, learning. The field is still solving medical problems; it is simply solving them with a different toolkit—more computational, more molecular, more automation-ready.

To a layman: medicine stops sprinting after a wildfire and goes back to urban planning. The heat of COVID recedes. Cancer, AI, healthcare systems, and computational biology take over the expensive front row.

Representative titles 2021: Mechanism of SARS-CoV-2 polymerase stalling by remdesivir
2025: The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies
Act III — Two ways to change

Chemistry walks. Engineering pivots.

Chemistry makes a visible move from a reaction-and-catalysis neighborhood into a frameworks-and-sensing neighborhood. The dominant cluster flips from 11 to 7. The vocabulary turns from reactions, catalytic, functionalization, methods, photochemical to sensors, analytical, frameworks, metal-organic, nanomaterials.

Engineering is the more subtle story. Its centroid only shifts 0.80 units. But its mix underneath changes: the battery-and-electrochemistry cluster rises from 16.8% to 24.4%, while the AI-and-vision cluster falls from 16.5% to 11.0%.

That is the hidden pattern of this map: some fields move by marching to a new neighborhood, and some move by quietly replacing the tenants while staying on the same block.

20212025Chemistry

Chemistry physically walks to a new neighborhood.

20212025Engineering

Engineering barely budges in space; the real motion is in who occupies the field.

Chemistry slides toward frameworks, sensing, and nanomaterials.
Engineering moves less in space than in composition. Batteries gain; AI-adjacent vision work loses share.
The twist

Computer science may be the field that changes the most while looking almost stable.

Its centroid barely moves compared with mathematics or medicine. But its language turns from learning, neural, network, few-shot, adaptation, domain toward systems, language, large, education, engineering, models. In other words: computer science does not leave its neighborhood. The revolution happens at home, inside the same district.

Dominant cluster stays the same: 0 Weighted shift: 0.63 AI/vision share falls while language-and-systems clusters rise
Reader mode

Pick a field. Watch the explanation change.

The big lesson of this corpus is that “shift” is not one thing. Sometimes it is a march across the map. Sometimes it is a collapse of one emergency cluster. Sometimes it is a quiet rotation inside the same neighborhood.

Shift summary

Mathematics

Mathematics shifts 6.56 map units, moving down-right. It begins in a neighborhood defined by methods, statistical, inference and ends near systems, optimization, education.

Mini trajectory
20212025
Representative titles

2021: Statistical Issues in Drug Development

2025: Dual approach artificial neural networks-fractional order operator to enhancing in vitro fertilization predictions and success measure

Terms that rise and fade
Green lines mark terms gaining share by 2025. Red lines mark terms fading from the 2021 frontier. Hover a line, dot, or label for start share, end share, and net change.
Cluster handoff

Dominant cluster: 10 → 4. Start terms: methods, statistical, inference, genomics. End terms: systems, optimization, education, engineering.

Caveat, honestly stated: this story tracks the top-cited papers in the OpenAlex sample, not the full universe of papers. The map axes are inferred from the dominant terms at each extreme, not officially labeled by UMAP itself. And 2026 is intentionally excluded because it is still a partial year. Even with those limits, the broad pattern is difficult to miss: the pandemic’s citation gravity weakens, and the new gravity wells are methods, machinery, materials, and AI-inflected biology.