What z0.0 is, and what it is not.

What it is

A single, harmonised catalogue of molecules reported in extracellular vesicles, assembled from the major public extracellular vesicle and proteomics repositories, with one set of counting rules applied across all of them.

You type a molecule. You get back how many published studies reported it, how many deposited samples it was measured in, which sample types and isolation methods and vesicle types those were, which species and which diseases, when it was first reported, and the list of papers behind every one of those numbers.

What it is not

• This release does not predict anything. Every figure on a molecule page is a count of something that was reported. No value on this site is inferred, modelled or estimated.
• It is not an abundance measure. A molecule reported in many studies is a molecule that has been looked at a lot. That is not the same as a molecule that is present at high levels, or one that matters.
• It is not complete. It covers what the public repositories contain. Work that was never deposited, or deposited somewhere not yet loaded, is not here. A molecule can be real and still be absent.
• It is not yet the model's predictions. VesTRA is a specialized foundation AI model, released in stages. z0.0 opens its foundation: the evidence layer the model learns from, every molecule, sample and study, counted one way. The model's own predictions arrive with z1, with a model card and an independent evaluation anyone can repeat.

Where the data comes from

This release is built from public sources. Each one keeps its own terms, and each should be cited alongside VesTRA when its data is used:

• Vesiclepedia (release 5.1), the community catalogue of proteins, RNAs and lipids identified in extracellular vesicles.
• ExoCarta (release 6), the catalogue of exosome cargo: proteins, RNAs and lipids.
• PRIDE, the PRoteomics IDEntifications database, at the level of individual deposited samples.
• jPOST, the Japan ProteOme STandard repository, for proteomics datasets deposited there.
• MassIVE and iProX, for further proteomics datasets deposited through ProteomeXchange.
• Panorama Public, for targeted proteomics datasets of vesicles.
• exoRBase (release 3.0), for circular and long RNAs measured in vesicles from human body fluids, sample by sample.
• exRNA Atlas, for small RNAs measured in vesicles from human body fluids, from the Extracellular RNA Communication Consortium.
• MetaboLights, for metabolomics studies of vesicles, with each study's own licence.
• Metabolomics Workbench, for metabolites measured in vesicles, with each study's own licence.
• Published supplementary tables, for patient-level measurements in vesicles that the papers themselves made public.
• ClinicalTrials.gov, for the registered clinical trials on extracellular vesicles.
• PubMed, for the title, journal, year and DOI of each study.

More repositories are being added. Each release says which sources it was built from and when it was built.

How the counting works

Counting sounds simple and is not. These are the rules, applied the same way everywhere on the site.

Molecule

One biological entity, not one name. The same protein appears across sources under a gene symbol, several accessions and a handful of historical names; those are resolved to a single entry, and the alternatives are kept as synonyms so searching any of them finds it.

Molecule type

What kind of molecule it is: a protein, a microRNA, a lipid, and so on. Each type is counted on its own, because adding proteins and RNAs into one number hides more than it shows. The total of all types is given as well.

Study

One published paper, or one public dataset that has no paper linked to it yet (for example a PRIDE or Metabolomics Workbench deposit). A dataset that has a paper is counted once, as the paper. A molecule reported many times inside the same study counts once for that study.

Sample

One deposited biological sample, identified by its source repository and its experiment identifier within that repository.

Measurement

One molecule measured in one sample, counted once. This is the largest of the four numbers and the rawest.

Every study links to its paper or, for a dataset with no paper yet, to its page in the repository it was deposited in. Placeholder identifiers are left out rather than guessed at.

Known limits of this release

• Coverage is uneven by design of the field: human plasma and cell-line work are heavily represented, other species and tissues much less.
• Disease and sample annotations are only as good as what the depositor recorded. Where a field was left blank in the source, it is blank here rather than filled in.
• Counts reflect reporting effort, not biology. Popular molecules look large because they have been studied, not necessarily because they are important.
• A number can change between releases as sources are corrected or added. Each page states the release it is showing.

What comes next

z0.1 helps you analyse and understand your own data. Bring the results of your own vesicle experiment, and VesTRA reads every molecule in it against the whole published record: what is well known, what is rare, what has never been reported, and what that means for your study.

z1 is the trained VesTRA model. Where z0.0 reports what has been measured, the model is being built to say which molecules a cell loads into its vesicles, and how that changes with condition. It will be released with a model card and an evaluation anyone can repeat.

Privacy

There is no account and no sign-in. You never have to say who you are.

To learn what researchers look for, the site counts visits anonymously: which pages are opened, which molecules are searched for, and how long a page stays open. Each browser gets a random code kept in that browser, never a name or an email. The hosting service also tells us the approximate country and city of a visit, and the network address is kept only as a one-way scrambled code that cannot be turned back into the address.

If your browser sends Do Not Track or Global Privacy Control, only plain page views are counted, with no code, no searches and no time on page. Nothing is sold, shared or used to build a profile of you.

---

Citing this release

VesTRA is a VUMENS resource. If you use it in work you publish, cite the release shown at the top of this page. A formal citation with a permanent identifier is being registered for this release and will be listed here.

Questions, corrections and missing sources are all welcome. Please report anything that looks wrong: a count that cannot be checked against its papers is a bug.