We were introduced to Dr Michael Castro by Stu’s oncologist, Dr Matt Williams. Dr Castro is a neuro-oncologist specialising in precision medicine and molecular oncology and is Chief Medical Officer at Cellworks. By the time we met him, we had already generated an enormous amount of molecular information about Stu’s tumour through CeGaT and Alacris. Dr Castro and Cellworks gave us another way of looking at that information.

Rather than simply identifying individual mutations and looking for a drug associated with each one, Cellworks used computational modelling to look at the interactions happening across Stu’s tumour biology and simulate how different treatments might affect it. For us, this was personalised medicine taken to another level.

WHAT IS CELLWORKS?

Cellworks is a precision medicine company that uses computational biosimulation to model an individual patient’s cancer and predict how it may respond to different treatments. Cancer biology is incredibly complicated. A tumour can contain multiple genetic alterations affecting different pathways, and those alterations don't necessarily behave independently of one another. That means finding a mutation and matching it with a drug isn't always as simple as it sounds.

Cellworks builds a personalised disease model using molecular information from an individual patient's cancer. Its platform then uses mechanistic biological modelling, biosimulation, AI and machine learning to model how that individual disease network may respond to different drugs and drug combinations. In very simple terms, instead of only asking:

“What mutations does this tumour have?” t can also ask: “What are those mutations doing together?” and: “What happens to this particular tumour model if we introduce this treatment or combination of treatments?”. That distinction was fascinating to us.

OUR EXPERIENCE

Dr Matt Williams introduced us to Dr Castro and we arranged a consultation. We sent Dr Castro the information we had accumulated about Stu's tumour, including the molecular data generated through CeGaT and Alacris. Cellworks then used Stu's individual tumour information to build a computational model. The resulting report was enormous. Around 180 pages enormous. It looked at the molecular pathways involved in Stu's tumour and modelled the predicted effects of different drugs and drug combinations. This wasn't simply a list of mutations followed by a list of drugs.

The modelling attempted to understand the relationships between different alterations within Stu's tumour and how those interactions might affect treatment response. The report gave us another layer of information to discuss with Dr Castro, Dr Williams and the other doctors involved in Stu's care.

DR CASTRO

One of the things we valued enormously about Dr Castro was his willingness to look at Stu as an individual. By this point we had become increasingly frustrated by the limitations of a one-size-fits-all approach to glioblastoma. Dr Castro's approach was very different.

He was interested in the molecular biology of Stu's cancer and what that might mean for Stu's treatment. That doesn't mean the computer could tell us exactly what would work. It couldn't. It didn't remove the uncertainty involved in treating glioblastoma and it didn't turn theoretical treatment options into proven ones. But it gave us another way of interrogating the information we had spent months gathering. And that felt incredibly valuable.

2026 UPDATE

Cellworks has moved on considerably Looking at Cellworks now, it's striking how much more developed the technology is than when we were navigating this with Stu. Cellworks currently describes its platform as a computational biosimulation platform powered by mechanistic biology models, biostatistics and machine learning. The platform creates a personalised disease model using multi-omic information from an individual patient.

It can then run virtual or in silico experiments to model how that patient's disease network may respond to different treatments. Cellworks says its current computational biology model contains thousands of human genes, hundreds of thousands of molecular interactions and models for hundreds of drugs. Those drug models can be combined computationally to investigate millions of potential treatment combinations. The platform currently provides two types of personalised therapy decision-support reports.

Singula predicts an individual's response to standard-of-care therapies. Ventura models combinations of approved drugs, including potential off-label combinations, for patients whose cancer is refractory to treatment. The aim isn't simply to identify a mutation and match it to a drug.

It's to model the biological consequences of the different alterations within an individual patient's disease and then simulate how treatments interact with that particular molecular network. Cellworks calls this personalised therapy biosimulation.

AND YES, AI IS PART OF IT. 

This is another example where AI isn't something we've retrospectively bolted onto Stu's story. Cellworks was already developing AI-informed biosimulation when Stu was being treated. Its current platform explicitly combines mechanistic biological models with AI and machine learning. I find that fascinating now because one of our biggest challenges in 2021 was trying to understand the extraordinary amount of information we had accumulated.

We had sequencing. We had mutations. We had pathways. We had reports. We had potential drugs. We had different doctors interpreting different pieces of information.

Cellworks was attempting to put those pieces together computationally rather than looking at each molecular finding in isolation. That feels even more relevant today than it did then.

WHAT ABOUT GLIOBLASTOMA? 

Cellworks continues to include glioblastoma within its clinical research. It has also published work using its biosimulation platform to investigate treatment response and resistance in GBM, including research into mechanisms that may influence response to temozolomide. That research is interesting, but as with all of the experimental and precision-medicine approaches discussed on this site, computational predictions and research findings aren't the same thing as proof that a particular treatment will work for an individual patient. The value for us was in generating more information to discuss with the clinicians treating Stu, not in treating a computational report as a prescription.

HINDSIGHT

WHAT WE KNOW NOW THAT WE WISH WE'D KNOWN THEN.

We had 180 pages. We didn't have 180 pages' worth of understanding. This is probably the perfect example of something I've talked about elsewhere on this site. By the end of Stu's treatment, we had accumulated an extraordinary amount of information. And the Cellworks report was extraordinary.

But there is a fairly substantial difference between being given approximately 180 pages of incredibly complex molecular information and actually understanding approximately 180 pages of incredibly complex molecular information.

We relied heavily on the doctors around us to interpret it. We tried to understand it ourselves. We Googled things. We cross-referenced reports. We researched drugs. We researched mutations. We tried to work out how the information from CeGaT related to Alacris, how Alacris related to Cellworks and what any of that might mean in the real world for Stu. It was overwhelming.

THIS IS WHERE I WOULD USE AI NOW. 

If I were handed that Cellworks report today, one of the first things I would do is use AI to help me interrogate it. Not to decide Stu's treatment. Not to override Dr Castro. Not to tell an oncologist that ChatGPT knew better. But to help me understand what I had been given. I would upload the Cellworks report alongside the CeGaT and Alacris reports and start asking questions.

What are the most significant findings across these reports? Where do they agree? Where do they appear to disagree? Which findings are simply descriptive and which might potentially have therapeutic significance? What evidence exists for a particular suggested treatment in glioblastoma specifically? Is the evidence clinical, observational, preclinical or theoretical? What does this pathway actually do? Why has Cellworks predicted a response to this particular drug? What questions should I ask Dr Castro about this finding? What questions should I take back to Stu's oncologist? That would have been incredibly useful.

Because the goal wouldn't have been to turn myself into Stu's oncologist. The goal would have been to walk into the conversation with his oncologist understanding enough to ask much better questions.

THE HUMAN STILL MATTERS

And strangely, looking at Cellworks today has reinforced that for me rather than changed it. Their technology is extraordinarily sophisticated. Their current platform can computationally model huge biological networks and millions of potential treatment combinations. But ultimately the information still has to exist within the context of a real person.

Stu wasn't a dataset. He was a man with glioblastoma. He had symptoms. He had previous treatments. He had medications. He had side effects. He had a tumour in a particular part of his brain. He had good days and shit days. He had things he was willing to try and things he wasn't.

A computational model can add another extraordinary layer of information. AI can help a patient or caregiver understand that information. But neither replaces the clinicians who understand the patient standing in front of them. For me, the sweet spot is all three working together. The technology. The medical expertise. And an informed patient or caregiver who feels able to ask questions. That's what I wish we'd had from day one.

CURRENT INFORMATION

Dr Michael Castro remains Chief Medical Officer at Cellworks. Cellworks continues to develop personalised therapy biosimulation using individual molecular data, mechanistic biological modelling, AI and machine learning. Its current personalised therapy decision-support products include Singula for modelling response to standard-of-care treatments and Ventura for modelling potential treatment combinations. Cellworks also continues to list glioblastoma among the cancer types involved in its clinical research. For current information about Cellworks, its technology, research and access to its personalised therapy decision-support services, please go directly to Cellworks.

dr michael castro & cellworks