A few weeks ago, I sat across from a biotech founder in the TDA studio and noticed a phrase he kept coming back to: "the average patient."

It sounds harmless enough. But in oncology, it shapes almost every treatment decision. A doctor weighs up the state of the cancer and a long list of other health factors belonging to the person in front of them, then prescribes what has worked best for the "average patient" whose profile most closely matches theirs.

Then comes the wait. It's often eight to 12 weeks before a scan reveals whether that drug was ever going to work on that specific tumour. Two to three months of side effects, hope and cost, for an outcome that is often positive, but sometimes tragically not.

Dr Christoph Meinert, co-founder and CEO of Brisbane-based biotech Gelomics, is trying to shrink that wait to something closer to zero. He grows a tiny stand-in for a tumour in a lab, tests treatments on that first, then gets Google AI to analyse the results. I sat down with him to find out how.

_{A note on this edition: This newsletter is sponsored by Google. As always, TDA has independently written and produced all editorial content for this newsletter without commercial influence.}

Dr (“Please call me Chris”) Meinert says Gelomics exists because the odds in cancer drug development are brutal.

As many as 19 in 20 cancer drugs that enter human trials never reach approval, according to a large MIT analysis of clinical trial outcomes. Getting one across the line in a successful trial can cost anywhere from roughly $650 million to $2.8 billion. I asked Chris why the odds were so stacked.

"The main problem is really that in drug development, many times good, well-working drugs are tested in humans without a proper selection of which patients would best benefit from this particular drug," he told me. "Often the drug is given to people where it won't show any effect, and ultimately the drug fails."

That means patients who might have benefited wait even longer for a treatment that's already sitting on a shelf, discarded for the wrong reasons.

Right now, Chris explained, doctors pick treatment using what he calls "a very rudimentary" method: cancer type and stage, plus a basic read of certain biomarkers in the tumour. Asked to put it in the simplest terms, he called it a flowchart.

Cancer doesn't behave like a flowchart, though. "We now understand that cancers are highly individual," he said. "One person's tumour is entirely different from another person's tumour."

That's the challenge Chris set himself: get doctors as much information as possible before a patient is exposed to a treatment at all.

This is where it gets sci-fi.

When Gelomics receives a small tissue sample from a patient's biopsy, they split it in two. One half goes toward digital spatial profiling, a cell-by-cell map of the tumour showing where the cancer and immune cells sit, and what each is doing. A single sample generates 30 to 50 terabytes of data. For scale, a 25-minute TDA podcast video is about two gigabytes, and there are 1,000 gigabytes in a terabyte. That's thousands of times more data, from a piece of tissue smaller than a fingernail clipping.

Gelomics uses the other half to grow tiny replicas of the patient's own tumour, called microtumours or patient-derived organoids. They're half a millimetre across, smaller than a grain of rice, but Chris says they closely mirror how the real thing behaves.

Once grown, researchers can test hundreds of existing and experimental drugs directly on the microtumour and watch the response before a single dose ever reaches the patient.

No human team could process this much information by hand. That's the job AI is doing here: crunching numbers, not dreaming up new compounds.

Gelomics trains AI models, built with Google's AI tools, on both the tumour map and the microtumour results, to predict how a specific patient might respond to a specific drug. Chris admits some of this works almost like a "black box." His team doesn't always know what pattern the model has locked onto, only that its predictions can be checked against how the real microtumour behaves.

That's an unsettling thing to hear about a tool involved in choosing someone's cancer treatment, so I pressed him on it.

The AI doesn't diagnose anyone, he said, and doesn't make the final call. "AI is really there to give a recommendation. It's only a recommendation. Ultimately, the decision lies with humans."

According to the Australian Institute of Health and Welfare, there were 170,000 new cancer diagnoses in 2025, or close to 470 people a day. Last year, cancer was responsible for roughly three in every 10 deaths nationally.

Chris wants a shift away from blunt, one-size-fits-all approaches and toward something built around what an individual's actual cancer looks like, not the average one.

There have been remarkable headlines lately about new cancer vaccines that could save lives outright. Chris isn't working on a cure. He's working on the way patients and doctors get information. But for someone staring down eight to 12 weeks of treatment, waiting to find out whether it was the right one, that could be one less guess they have to make.

Cancer treatment is usually designed for the "average" patient. What if it didn't have to be?

Cancer is deeply personal. Yet, because of how standard care works, doctors often have to rely on what succeeded for someone else in clinical trials - not what will work for your specific biology.

For some this trial-and-error approach means enduring weeks of exhausting therapy that was never going to work. Brisbane biotech Gelomics wants to give that time back.

Brisbane biotech Gelomics grows patients' tumours in the lab, then trains Google AI on how those tumours respond to real drugs. The goal is matching the right treatment to the right patient, and getting new cancer drugs to the people they'll actually help, faster. We spoke with CEO Dr. Christoph Meinert about the deeply personal future of care.