Taxonomy

OSCR covers neuroscience, and classifies each paper it reads along four facets: the kind of data it collects or analyses, the species studied, the condition studied, and its angle. The definitions below are the ones its classifier and its annotators use; the counts are those of the catalogue this site was built from.

Scope

The registry's field is neuroscience, defined so: Neuroscience means research on the nervous system (brain, spinal cord, nerves, retina: their cells, circuits, development and diseases), or research on the mind that uses neural measures (recordings, imaging, stimulation, lesions, neural tissue). Psychology or psychiatry with questionnaires and behavior only is not neuroscience; neither is a paper where 'neural' only means an artificial neural network.

In scope

The paper studies the nervous system, or studies the mind with neural measures. For example: an fMRI study of working memory in healthy adults; dopaminergic neuron loss in a mouse model of Parkinson's disease; a trial of an antiseizure drug in focal epilepsy.

Out of scope

Anything else, including psychology or psychiatry without neural measures and machine learning outside neuroscience. For example: a questionnaire survey of stress among university students; a convolutional neural network that grades colon biopsies; cortical bone screws in hip surgery.

How papers come in. The harvester asks Europe PMC for the open-access papers whose full text it holds and whose title or abstract names a neuroscience term — EEG, MEG, fMRI, intracranial recordings, spike sorting, calcium imaging, electrophysiology, brain–computer interfaces, neuroimaging, fNIRS, transcranial magnetic or deep brain stimulation, neural oscillations, the connectome, diffusion MRI, neurons, neural, cortex, cortical, hippocampus, brain. The query is broad on purpose, and catches papers that are not neuroscience ("neural network", "cortical bone"): the classification then decides.

Why off-topic papers stay out. A paper judged out of scope stays on the operator's computer: it has no page, is in no list, search, lookup or dataset, and counts in no figure. Harvesting broadly and filtering after keeps the registry complete for its field without letting unrelated papers dilute its rates — a decision of 27 September 2026. 22,128 papers read are in scope today.

Which papers have a page

Among the papers in scope, those with something to show about code or data have a page; every other one is known to the DOI lookup.

4,974 papers have a page, 2,882 of them with their authors' code. What each status means.

How to read the categories

Each facet has a fixed list of values. A value has a key (modality.eeg), a name, a definition and examples, and the number of papers with a page it applies to (with, in brackets, those with their authors' code); its name leads to its page on Browse when it has papers. Modality, organism and population take several values per paper; the subfield takes one. The vocabulary's version: 2026-09-27.

Modality

The kind of data the paper itself collects or analyses (for a review: the kind of data it discusses). List every modality that is a real part of the work, not those only mentioned in the background. A paper may have several.

Organism

The species studied: where the subjects, tissue, cells or data come from. For in vitro work, the species the cells come from (human iPSC-derived neurons are human). A paper may have several.

Population

The condition studied. 'healthy' when the study is about typical subjects only (healthy people, wild-type animals, healthy aging); otherwise the condition(s). Healthy controls in a patient study do not add 'healthy'. An animal model of a disease counts as that disease. Write '-' when there are no subjects at all (pure theory, a simulation, a method on synthetic data). A paper may have several.

Subfield

The paper's main angle. Pick one; when two fit equally well, pick the one that best describes the main question and name the other in the notes. One per paper.

How a paper gets its categories

1. Rules first. Weighted patterns run over the paper's title (strongest), keywords, MeSH terms, subjects, abstract and journal; its type and methods add weak hints. Every value comes with its confidence and the reasons that fired ("title: 'EEG'"). When the rules find nothing, or two values compete closely, the facet is marked ambiguous rather than guessed.
2. The operator's labels win. Where the operator has labelled a paper by hand, the label overrides the rules; a sample of such labels serves to measure them.
3. A local model for the ambiguous cases — once one has been compared with the operator's labels and chosen. It runs on the operator's computer, and uses its graphics processor only between 01:00 and 07:00. None is in use yet.
4. What is shown. A category appears on the site when the operator or a model set it, or the rules did with a confidence of at least 60% and no ambiguity. Otherwise the paper is simply not in that category: an unclassified paper is still in the list and the search.

OpenAlex's topics

OpenAlex, an open index of research, gives each paper a primary topic within its own hierarchy — domain, field, subfield, topic. OSCR shows it on each paper's page as OpenAlex gives it; it is not the registry's classification. Among the papers with a page, 4,937 have one:

Types of papers

The papers read in scope, by their type as the publisher declares it. The registry's rates — the share of papers with their authors' code — are computed on research articles only (research articles, brief reports, methods articles, data papers, rapid communications, short reports): reviews, conference abstracts, case reports and notices rarely have code of their own.