Yes, for runs the lab owns outright. Buyers ask life-science labs for structured run records with measured results, failed runs included, that have not been published. The unit that carries value is the raw file together with its method and the outcome it led to. Runs made for a sponsor usually belong to the sponsor. Human sequencing data and images tied to patients are personal data and depend on the consent they were collected under.
- The raw file, the method and the outcome together are the unit. One without the others is weak.
- Sponsor-funded runs usually belong to the sponsor, even when the CRO keeps the archive.
- Failed runs are wanted if they are labelled and linked to what happened next.
- Human genomic data follows the consent it was collected under.
Can a lab license raw instrument data to AI buyers?
Yes, when the lab owns the runs and they have not been published. Chromatograms, spectra, micrographs and reads are licensed as part of run records: the raw file plus the method that produced it and the result it led to.
A public listing for biotech and life-sciences labs (one buyer's public listing, read 15 November 2026) asks for "life-sciences R&D organizations whose experiments are written down: lab notebooks, structured run records with measured results, and clinical trial documentation." Its requirements include "Electronic lab notebooks or structured experiment records the organization owns", "1,000+ runs with their inputs and measured outcomes, failed runs included" and "Data that hasn't been published".
It names who it has in mind: "Biotech and biopharma, CROs, genomics and diagnostics labs, consumer genomics companies, academic spin-outs". And it says what does not fit: "Data that has already been published" and "Data owned by a sponsor or partner".
Why raw instrument files are scarce
A paper shows one chromatogram, cropped, and a table of the numbers that mattered. The hundreds of injections behind it stay on the instrument PC or a network share, in the vendor's own format, and are rarely seen by anyone outside the lab.
Regulated labs already treat those files as the record. Good laboratory practice defines raw data as the "laboratory worksheets, records, memoranda, notes, or exact copies thereof, that are the result of original observations and activities of a nonclinical laboratory study", and says it "may include photographs, microfilm or microfiche copies, computer printouts, magnetic media, including dictated observations, and recorded data from automated instruments" (21 CFR 58.3).
That is the scarcity. Processed results are public in the literature. The original signal, with its baseline drift, carry-over, failed system suitability and re-injections, is held by the lab that ran it.
Method plus result plus outcome: the unit buyers pay for
A raw file on its own cannot be interpreted. Each instrument has a small set of companions that turn a file into a record.
| Instrument | Raw file | Method to keep with it | Outcome that gives it value |
|---|---|---|---|
| HPLC and UPLC | Detector trace per injection, sequence file | Column, mobile phase, gradient, flow, temperature, integration events | Purity or assay result, pass or fail against the specification, the investigation if it failed |
| Mass spectrometry | Vendor raw file with all scans | Source and acquisition settings, calibration, sample preparation | Identified compound or peptide, quantity, whether the hit was confirmed |
| Microscopy and imaging | Full-resolution image stack | Objective, channels, exposure, stain or label, pixel size | Scored phenotype, pathologist or scientist call, the decision it informed |
| Sequencing | Reads with quality scores | Library preparation, platform, run parameters, reference used | Variant calls or expression table, validation result |
| Plate readers and assays | Well-level readings per plate | Plate map, reagents and lots, incubation, controls | Curve fit, hit call, whether the result repeated |
The sample ID is the thread. If the ID in the instrument file matches the ID in the notebook and the LIMS, the three records join. If it was retyped by hand each time, expect to spend time on a mapping table.
Can a CRO license instrument data from client studies?
Usually not without the sponsor. Study data produced under a services agreement normally belongs to the sponsor, and the listing treats "Data owned by a sponsor or partner" as not a fit. A CRO's own method development and internal validation runs are a different matter.
Keeping the archive is not owning it. Good laboratory practice requires that "documentation records, raw data and specimens pertaining to a nonclinical laboratory study and required to be made by this part shall be retained in the archive(s)" for set periods (21 CFR 58.195). That is a duty to keep, and it says nothing about who may license the content. The master services agreement does.
Grant-funded work has a written rule. "Title to intangible property acquired under a Federal award vests upon acquisition in the recipient or subrecipient", but the recipient "must not encumber the property without the approval of the Federal agency or pass-through entity" (2 CFR 200.315). Read the award terms. The guide for biotechs, CROs and contract labs covers sponsor agreements in more detail.
Are failed runs worth keeping?
Yes. The listing asks for "failed runs included" and says "Failed experiments are as valuable as successful ones." A failed run shows what a problem looks like in the signal, which a published result never does.
Two kinds of failure are worth telling apart. An instrument failure, such as a lost prime, a blocked needle or a failed system suitability test, teaches a model to recognise a bad run. A scientific negative, such as a compound that did not bind, teaches it what does not work. Label which is which.
A failed run is usable when the failure is recorded and linked to what came next: the investigation, the corrective step and the repeat injection. A folder of unexplained aborted runs is noise.
Does sequencing data count as personal data?
Human sequencing data does. US health privacy rules treat genetic information as health information, EU law puts genetic data in a special category, and research consent forms often limit reuse. Sequence data from cell lines, microbes, plants and animals does not carry those limits.
Under the US health privacy rules, "Health information means any information, including genetic information, whether oral or recorded in any form or medium" (45 CFR 160.103). The research rules define an identifiable biospecimen as one "for which the identity of the subject is or may readily be ascertained by the investigator or associated with the biospecimen" (45 CFR 46.102), and a consent form for research on biospecimens has to say "whether the research will (if known) or might include whole genome sequencing" (45 CFR 46.116).
EU law defines genetic data as "personal data relating to the inherited or acquired genetic characteristics of a natural person which give unique information about the physiology or the health of that natural person", and Article 9 says the processing of genetic data "shall be prohibited" unless an exception such as explicit consent applies (GDPR).
The practical test is the consent form. The listing's own wording is "Genomic data collected with research consent", and its scoping question is "Is any of it patient-level or genomic data, and under what consent?" If the form does not cover commercial reuse or sharing with third parties, the human data stays out. This page is general information, not legal advice.
Checklist: file formats and metadata to keep
No public listing read on 15 November 2026 gives a list of accepted instrument formats. The safe course is to keep the vendor raw file and add an open export next to it, with the method and audit trail.
Open formats exist for the main instrument types. For mass spectrometry, the standards body "developed a new format called mzML by taking the best aspects of each of the precursor formats to form a single one" (HUPO PSI). For microscopy, OME-TIFF "takes advantage of the rich metadata defined in OME-XML while retaining the pixels in multi-page TIFF format" (OME). For sequencing, the public archive "accepts binary files such as BAM, SFF, and HDF5 formats and text formats such as FASTQ" (NCBI).
- The vendor raw file, untouched, for every injection, scan or plate.
- An open export beside it: mzML for mass spectra, OME-TIFF for images, FASTQ or BAM for reads, a text or CSV trace for chromatograms.
- The acquisition method and the processing method, as files, not as a description.
- The sequence or worklist that shows run order, blanks, standards and controls.
- Sample IDs that match the notebook and the LIMS.
- Calibration and system suitability results for the day.
- The audit trail. Regulated systems keep "secure, computer-generated, time-stamped audit trails" of entries and changes, and it shows which result was reprocessed.
- Instrument model, software version and column or consumable lot.
- The result table and the decision: pass, fail, repeat, investigate.
The audit trail wording is from 21 CFR 11.10. Export while the instrument software licence is still active. Old vendor formats are hard to open once the workstation is gone.
How many instrument runs make a dataset?
One public listing sets the bar at "1,000+ runs with their inputs and measured outcomes, failed runs included". A busy analytical lab passes that in weeks, so the count is rarely the problem. The share of runs that still have a method and an outcome attached usually is.
Instrument data is the middle layer of three. The notebook says why the run was done, the instrument file holds what was measured, and the LIMS holds the sample and the reported result. ELN vs LIMS explains where each record sits, and the R&D experiment records guide covers the notebook side. Records kept under good laboratory practice follow the rules in the 21 CFR part 58 guide.
Example: one HPLC run, assembled as a record
A fictional analytical lab assembles one stability-study injection that failed and was repeated.
| Piece | Record | Where it was |
|---|---|---|
| Why | Notebook entry: 6-month stability pull, lot 14, assay and impurities | Electronic lab notebook |
| Method | Acquisition and processing method files, version 3 | Chromatography data system |
| Raw signal | Detector trace for six injections with the sequence file | Chromatography data system |
| Failure | System suitability failed on tailing factor; audit trail shows no reprocessing | Chromatography data system |
| Investigation | Deviation record: column at end of life, replaced | Quality system |
| Repeat | New sequence, suitability passed, assay 98.7 percent | Chromatography data system |
| Reported result | Sample result approved and released | LIMS |
Fictional example. Seven linked items from four systems. The failed first attempt and its investigation are the part a published stability table never shows.
Common mistakes with instrument data
Most of them happen at clean-up, when a project closes or an instrument is retired.
- Exporting PDFs of reports and discarding the raw files.
- Keeping raw files without the methods that produced them.
- Deleting failed and aborted runs to tidy the archive.
- Counting sponsor studies as the lab's own data.
- Mixing human samples into a set without checking the consent forms.
- Letting the instrument software licence lapse before the export.
- Retyping sample IDs so that instrument, notebook and LIMS no longer match.
Instrument data as a data asset
One published company range applies (one buyer's public listings, read 15 November 2026). The listing for biotech and life-sciences labs runs from $50,000 to $875,000 and says the figure "Depends on volume and depth of records". Read it as a published ceiling, not a typical outcome: it assumes owned, unpublished runs with inputs and outcomes attached.
A lab that has closed a programme or is winding down still has the files, and whoever holds title can license them. The description says "Biotechs that have wound down and labs with years of archived work are a strong fit." There is no standard timetable. The order is usually the same: a written description of the archive, a small sample, a rights review, agreed terms, then delivery. The rights review sets the pace. Manufacturing-side records are covered in the GMP batch records guide and chemistry and materials work in the chemistry and materials R&D guide. Origin Data Partners is an independent data partner that helps companies get their data bought. A buyer may pay Origin when a deal closes.
Does your company hold records like these? See what your data is worth.
Common questions
Can a CRO license instrument data from client studies to AI companies?
Usually not without the sponsor. Study data made under a services agreement normally belongs to the sponsor, and one public listing treats sponsor-owned data as not a fit. A CRO can license its own method development and internal runs.
Are failed HPLC or mass spec runs worth keeping for AI licensing?
Yes. One public listing read on 15 November 2026 asks for failed runs to be included. They are usable when the failure is labelled and linked to the investigation and the repeat run.
Which instrument file formats do AI buyers accept?
No public listing gives a format list. Keep the vendor raw file and add an open export: mzML for mass spectra, OME-TIFF for images, FASTQ or BAM for sequencing reads and a text trace for chromatograms, each with its method file.
Does sequencing data count as personal data under HIPAA and GDPR?
Human sequencing data does. 45 CFR 160.103 includes genetic information in health information, and GDPR Article 9 puts genetic data in a special category. Reuse depends on the consent it was collected under. Non-human sequence data is not personal data.
How many instrument runs make a licensable dataset?
One public listing sets a threshold of 1,000 or more runs with their inputs and measured outcomes, failed runs included. Most labs pass the count easily. The harder test is whether each run still has its method and outcome attached.
Can you sell microscopy images for AI training?
Yes, if the lab owns them and they are not tied to identifiable patients. Keep full-resolution stacks with objective, channels, stain and pixel size, plus the scored result. Images from human tissue follow the consent and privacy rules for that sample.
Sources and further reading
- 21 CFR 58.3: definitions (raw data)
- 21 CFR 58.195: retention of records
- 21 CFR 11.10: controls for closed systems (audit trails)
- 2 CFR 200.315: intangible property under a federal award
- 45 CFR 160.103: definitions (health information)
- 45 CFR 46.102: definitions (identifiable biospecimen)
- 45 CFR 46.116: general requirements for informed consent
- Regulation (EU) 2016/679 (GDPR), Articles 4 and 9
- HUPO Proteomics Standards Initiative: mzML
- Open Microscopy Environment: the OME-TIFF format
- NCBI Sequence Read Archive: file format guide
