Can a Biotech Sell or License Its Failed Experiments to AI Companies?
AITrainingDataBroker.com
Can a Biotech
Sell or License
Failed Experiments?
3 parts: what a notebook holds · how AI learns from it · who may license it
Jimmy Lin
Can a Biotech Sell or License Failed Experiments?
In a biotech notebook, the run that failed sits right next to the run that worked.
That page is what this video is about.
So can a biotech sell or license its failed experiments to AI companies?
The video
Full script
In a biotech notebook, the run that failed sits right next to the run that worked.
That page is what this video is about.
So can a biotech sell or license its failed experiments to AI companies?
A report by BIO, a biotech trade group, looked at drug programs that reached a first human trial in 2011 to 2020.
Only about 8 in 100 of them won approval.
So most of a biotech's work ends as notes about what did not work.
Those notes often sit in a drawer or on an old server.
This is general information and not legal advice.
First, let me tell you who I am.
I'm Jimmy. I have worked in data at Fortune 500 companies, as a data analyst, a data engineer and a software engineer, and I am trained in data science. I am also a Canadian military veteran, specialized in data analysis and reporting.
If you want to sell your data, I am your broker and technical advisor. I bring your data to several AI companies, and I guide your team through the technical side, including compliance. AI companies pay my fee.
Already shopping your data around? Contact us anyway. We can still work through the terms with you and help you negotiate a better price.
If you are buying data for an AI lab, I find the right data and see it through to delivery, so you can focus on AI research.
Now, what is actually on a notebook page?
A chemist writes down what was tried, what came out, why it went wrong, what changed, and what the next run will be.
The failure and the fix sit side by side, which a published paper almost never shows, because a paper keeps only the run that worked.
That pairing of settings and results is the whole point.
Most people think a lab keeps only what worked.
It does not.
A lab keeps the misses too.
And a miss written next to its settings shows something a clean success can not: which setting changed the result.
Here is a real case.
In May 2016, the journal Nature published a study by a team of chemists and computer scientists.
They trained a computer model on failed reactions taken from archived lab notebooks.
In their tests, the model beat human strategies at predicting which conditions make new crystals form.
So the failed runs were not waste, they were the training data.
What do AI companies ask for when they look at notebook text? The requests tend to look alike.
They want the whole process, not only the results.
Each entry should say success, failure or inconclusive, and name the field, such as organic chemistry.
They also ask where it came from: the company's own files, a government lab archive or a university archive.
And they want proof that personal information was removed.
The text goes as PDF or plain text, and the figures come as separate image files.
Some AI companies also ask for a mix of languages.
Next comes the question that stops most deals: who has a say?
In one notebook there can be the company, a sponsor that paid for the work, and a partner lab or university.
Each of them can hold rights.
So your counsel reads the research agreements first, before anything is prepared.
If a notebook touches patients or trial participants, those entries are not licensed unless the consent allows it and the personal information can be removed.
This will not fit every notebook, and that is fine.
When health information is in a notebook, the US health privacy rule, called HIPAA, gives 2 ways to de-identify it.
In the first, an expert studies the data and writes down why the risk of identifying anyone is very small.
In the second, you remove 18 kinds of identifiers, such as names, small places, dates, phone numbers and record numbers.
The rule is 45 CFR 164.514. It covers health information held by providers and insurers, so a biotech may fall under other rules, but it is a common yardstick.
And consent still comes first, because what participants agreed to sets the limit.
Here is a different question: how long must a lab keep its data?
For work paid by a US federal award, rule 2 CFR 200.334 says to keep the files for 3 years after the final financial report.
It says longer if a claim or an audit is open.
For lab studies under the FDA's good laboratory practice rules, 21 CFR 58.195 sets a few periods.
One is at least 2 years after the FDA approves the application, and another is at least 5 years after the results are submitted.
The shortest one that applies is used.
But keeping a file is not owning it, and none of these rules says who may license the notebook.
Suppose a small biotech logs its runs for 4 years, failures included.
In year 5 its lead drug fails and the company closes, and the notebooks go into storage.
Later, someone who can sign for the company checks who owns each study, and only the work the company owns outright is licensed.
A closed company can still qualify, because the data is still there, as long as someone can sign and the rights are clear.
This company is not real, it is only here to show the order of the steps.
What about your own data? What would an AI company ask about it first?
There are 5 questions.
Are the failed runs still there?
Is each entry tied to its result?
Who owns each study?
What did a sponsor's contract say?
And is there patient data inside?
If you can answer those, you already know where you stand.
How do you answer those 5 questions for your own notebooks? Most owners do not have to do it alone.
A deal usually runs the same way. A small, de-identified sample goes to several AI labs, each one makes an offer, and you compare them and make the final call.
That is the part I do with you. As your broker, I bring back the offers and work through the terms with you. As your technical advisor, I guide your team through the export, the format, the de-identification and compliance.
We move forward only after your team signs off, and the AI lab usually pays my fee, after the deal closes.
A failed experiment is not waste: a notebook that keeps the miss next to its settings may be worth licensing, once the rights are clear.
More on how licensing works is at AITrainingDataBroker.com.
Which failed experiment in your lab would you most want a computer to learn from? Tell me in the comments.
Thank you for watching, and I will see you next time.
More slides
AITrainingDataBroker.com
AI Labs Want
Lab Notebook Data
Right Now (Nov 4)
An open request for lab notebook data, what it asks for, and what to do today
Jimmy Lin
AITrainingDataBroker.com
Has Your Data Been
Licensed Before?
AI Labs Ask First
Why an earlier license changes what your data is worth, and how an AI lab checks
Jimmy Lin
Sold out, then thrown out
