You Never Agreed To Be Studied

There are some rare genetic diseases where researchers know the exact mutation responsible for the condition, but developing a treatment has never been commercially realistic. With only a handful of those affected, it's hard to justify the cost of a conventional drug development program. So we find ourselves in this unfortunate situation where we understand the underlying biology but have no medical treatments to offer.

That’s why I found the FDA announcement this week to be encouraging as the agency wants to rethink how treatments for rare diseases with very small patient populations are developed and evaluated.

Right now, only about 5% of rare diseases have an FDA-approved treatment. If AI eventually makes it cheaper to design therapies, and regulators find workable ways to assess them, then maybe some of those affected by the rare disease will finally get options.

So overall, this version of personalized medicine is becoming real, giving hope to those where really had no choice but to live with the disease. This is all great news, but is also where I’d want to open up the discussion of personal privacy as we head into this future.

There’s something about a treatment developed for one person that I hadn’t thought much of before. In a trial involving thousands of patients, you can publish results without making any one participant recognizable. But in a one-person study, the paper may need to describe the exact mutation and response over time. If only one family is known to have that condition, taking out the patient's name might not accomplish much.

Of course, the FDA isn’t asking researchers to disclose identities, and sensitive genetic information can be kept in a controlled-access databases.

Still, if another child with the same mutation comes along fifteen years later, their doctors would probably want to know how the first patient’s treatment held up, including any late complications. Meanwhile, that first patient might be an adult who doesn’t particularly want to remain a research subject for the rest of their life. This should be their choice, no? Are they burdened by the obligation of follow-up tests?

And if we take this a bit further, it might not stop with the patient. After all, relatives share a lot of DNA. Suppose a child’s parents agreed to share detailed genomic data because it helped researchers develop a treatment. Twenty years later, AI systems may be able to identify health risks in those same sequences that nobody understood back in 2026.

Depending on the inheritance pattern, those new predictions could have implications for the patient's siblings or future children, even though they never agreed to participate!

You see, this is where we could cross some very questionable ethical lines. What unsettles me is that the family might have made a perfectly reasonable decision to share information for a lifesaving treatment, and the researchers might have handled it carefully. It’s just that genetic information doesn’t stay scientifically frozen. As our models improve, the same old data may begin to mean something quite different.

It reminds me of the film Gattaca (great movie, very relevant!), where someone’s future is judged by their DNA. Except our version could involve people who have already been treated successfully, or relatives who have never been sick.

Imagine applying for life insurance in 2046 and discovering that a prediction about your health was partly influenced by a cousin’s childhood medical record. In the US, federal law protects against genetic discrimination in employment and health insurance, but its main protections don’t extend to life, disability, or long-term-care insurance.

Insurers don't automatically get access to confidential research data, of course, but I can see how the boundaries might become blurred.

I really hope personalized medicine becomes an everyday reality for rare diseases, because it’s long overdue. But I wonder if, while we’re finally learning to treat people one at a time, we might also be creating risk predictions about themselves or relatives years later without anyone quite intending to.

I looked through our database of AI drug discovery programmes and found 36 rare genetic diseases being targeted. What’s interesting is that 25 of them are inherited, which means the genetic information from one patient could also tell us something about their family.

For example, if a child is sequenced to develop a personalized treatment, that information might reveal that one of their parents carries the same mutation, or help predict the chances of their siblings or future children inheriting it.

Now, I can understand why parents would be willing to share this information in 2026 if it could help save their child's life. What I’m less sure about is what happens twenty years later, when better AI models can revisit the same genetic data and identify health risks nobody knew about at the time.

So information originally collected to treat one child could eventually affect other members of the family who were never part of the study!

Google announced $36 million in funding for 17 organizations using AI to advance scientific research. Projects range from cancer treatment and drug discovery to genetic research and climate science. Among the recipients is the University of Toronto, which is developing an AI model to help design peptide-based medicines. [Link]

The UK's medicines regulator unveiled plans to use AI to help review clinical trial protocols. As part of its 2030 strategy, the MHRA plans to develop an AI-powered tool to check trial applications for potential issues before submission. It also plans to publish an AI Medicines Development Action Plan in 2027 and establish a dedicated Centre for AI and Digital Health. [Link]

David Baker's team published two Nature papers showing that AI can design enzymes that cut proteins at specific locations. In one study, 36% of 135 designs successfully cut their intended peptide targets, and researchers demonstrated activity against proteins associated with Alzheimer's and other diseases. A second study showed that newly designed enzymes could also work inside cultured human cells. [Link]