Why is biological data so scarce?
Ask a biotech founder what’s slowing them down and you’ll rarely hear “we’ve run out of ideas.” The bottleneck is almost always data. High-quality biological datasets are scarce for reasons that don’t go away with a bigger budget: experiments take months to run, cost a fortune, and are unforgiving: one off variable and an entire batch is wasted. The data that would tell you whether an idea works is precisely the data you don’t have yet, and can’t cheaply get.
Layer on top of that the data you’re not allowed to use. Patient records, clinical measurements and anything touching a real person is locked behind privacy law and ethics approval, for good reason, but with the side effect that the richest datasets are often the least accessible.
What does the data gap actually cost you?
It costs you in three currencies. Time: months waiting on experiments or data access, while your runway shrinks. Money: R&D budget burned on physical iterations that a smarter process would have skipped. And confidence: models trained on datasets that are too small, too skewed, or too narrow to trust, shipped anyway because there was nothing better.
The quiet killer is the third one. A model that looks great on a small, clean dataset and then falls over on the messy reality of the lab doesn’t just waste the work that went into it. It erodes trust in the whole approach.
Can’t you just throw more AI at it?
This is the seductive wrong answer. AI models are powerful, but they’re only as good as the data they’re grounded in. A sophisticated model trained on thin or biased data doesn’t fix the data gap. It launders it, turning “we don’t have enough data” into “we have a confident-looking model that’s wrong at the edges.” More parameters don’t manufacture evidence that was never collected.
So the real problem isn’t a shortage of algorithms. It’s a shortage of trustworthy data. Which means the solution has to operate on the data, not just the model.
Your real options: brute force, black box, or a virtual experiment
Faced with the data gap, teams usually reach for one of two bad options:
- Brute force. Spend your limited budget running more physical experiments, hoping you stumble onto the right configuration. Thorough, but slow and expensive; your budget runs out long before the search space does.
- Black box. Lean on off-the-shelf AI that performs well in a paper and falls over in the lab, because it was never grounded in the physics or biology of your actual system.
There’s a third option that’s easy to miss: don’t wait for the data, make it. If you can’t collect the data you need, you can either simulate it or synthesise it.
Manufacture it or simulate it: how synthetic data and digital twins fit together
These are the two honest ways to close the data gap, and they solve different halves of the problem.
Synthetic data manufactures realistic datasets. You learn the structure and statistics of the real data you do have, then generate new data that carries the same signal (the same patterns, correlations and edge cases) without containing a single real record. It’s how you get volume where real data is scarce, balance where it’s skewed, and privacy where it’s sensitive. It answers: “give me data that behaves like the real thing.”
A digital twin simulates the system itself. A physics-anchored model lets you run experiments virtually, thousands of conditions in seconds, and tells you with a confidence score which results to trust and which to verify. It answers: “what would happen if…?”
They compose. A twin can generate synthetic data for conditions you’ve never physically tested. Synthetic data can seed and validate a twin. Used together, they turn the data gap from a wall into a workflow: simulate to explore, synthesise to train, and spend your precious wet-lab time only on the experiments that actually earn it.
The data gap isn’t going away. Biology will always be slower and more expensive than software. But it doesn’t have to be the thing that sets your pace. Stop waiting for the data. Start making it.
Two engines for one problem: explore the ALTDATA Digital Twin and Synthetic Data.