How tech is powering precision treatment
While medical treatment has made leaps and bounds throughout the years, and continues to amaze and realise the impossible as we move forward, the sheer breadth of human diversity has always proved a key challenge. What works for one, does not necessarily work for another, and it is this fundamental issue, evidenced constantly in everyday treatment, which continues to drive medical consensus toward the need for more personalised, precision approaches.
To this end, Pharmafocus spoke to Dr Olivia Rossanese, Head of Biology in the Cancer Research UK Cancer Therapeutics Unit within the Division of Cancer Therapeutics at the Institute of Cancer Research (ICR), to determine exactly what we mean by the term “precision medicine”. She explained:
“Precision medicine can be broadly defined as ‘getting the right drug to the right patient at the right time.’ When you describe it that way, that doesn’t sound that different to what doctors have already been doing, but what I think has changed is the level to which we can understand a patient’s disease. So when we say ‘the right patient’, we’re talking about getting all the way down to the molecular level; especially for something like cancer, you can sequence a patient’s tumour to understand any genetic mutation, or look at protein or RNA expression to define a molecular signature.
“We’re trying to bring things closer together so now when we talk about the right patient and the right drug, we’re making them a lot more specific, based on a lot more data,” she added.
The importance of new technologies in this pursuit cannot be overstated. As our knowledge grows and we uncover ever more variables about patients, their conditions, and the therapies used to treat them, we are able to build more robust treatment methodologies, specifically tailored to the individual. But only if we can effectively handle the increasingly vast amounts of clinical data generated – something the human mind alone cannot do. Artificial intelligence, machine learning, and other technologies, when applied effectively, can help overcome these challenges and pave the way to better outcomes for individual patients.
This importance is being increasingly recognised. Recently, the UK Government has committed to investment in the value of such technologies, revealing plans to launch five specialist collaborative centres across the country which will utilise advanced digital technology including AI to push for earlier diagnosis and better outcomes, particularly in the treatment of cancer.
With this in mind, Dr Rossanese expanded upon the interwoven relationship between technology and precision medicine, and how the former enables the latter by allowing researchers to handle and analyse vast amounts of data on a scale just not possible via other means.
“On the first level – getting the right drug to the right patient – a lot of what we think about precision medicine is already deeply ingrained in complex data analysis and machine learning,” she explains. “Sequencing the patient’s DNA and looking at the proteins that are expressed and all of that – all of this creates huge amounts of data, which has to be meaningfully integrated in a way that I can use it, because I can’t keep fifteen different variables about this patient in my head all at once. You can use data pipelines to create this molecular signature, which you can then compare with other patients’ molecular signatures and compare with response.
“A lot of this is already going on; the next level to that is big data,” she continues. “If we agree that heterogeneity is the problem, you are actually talking about a large amount of data even for one patient and one tumour and one treatment. You’re also talking about a lot of different sets of data. Mutation data is quite binary – either you have a mutation or you don’t. Protein expression data happens on a scale from very low at zero to very high at a thousand, and then outcome data can be scribbled down by clinicians on a piece of paper. You need very large data, you need different data, and you have to integrate it in order to get the picture that you need. It’s a challenge to our current ways of handling and analysing data, which is where machine learning comes in. Machines themselves will learn how to parse this data and how to integrate it; once it has all of the data it can make a connection that we can’t see, because we can only hold a few things in our head at one time, and we haven’t previously been able to look at the totality of all this data.”
In addition to allowing for the simultaneous aggregation of data with the end goal of providing more precise, tailored treatment for patients, these technologies can be applied at the earliest end of the clinical development cycle and have transformed the way that we generate new medicines. This allows for a much more methodical approach than was ever possible previously, as Dr Rossanese details:
“The concept of precision medicine has huge implications for how we approach drug discovery and development. If you think about something like Aspirin, it was identified in a phenotypic screening, and it seemed to stop people having pain, but we had no idea why; it got approved and lots of people used it. Now, we’re not doing that anymore; it’s no longer a case of identifying the target and making a drug against it, which in itself is challenging. It’s about developing a therapeutic hypothesis about which patients will respond, so at the same time you’re doing the drug discovery, you’re doing the medcam to actually generate the molecule – you should be building this preclinical evidence for that hypothesis to understand which patients will respond and why. Also, you’re finding a way to identify those patients. So ideally you’re delivering a drug, a biomarker that shows that you can measure that patients are responding, and a diagnostic that helps you find those patients in the first place. It’s a more holistic package of data that you delivering, and not just ‘hey, Aspirin looks like it works.’
“That’s a lot of extra work; however, it’s a lot more information, so you understand a lot more about what your drugs are doing. The benefit of that is that if you have hypotheses about the proper patient population, you can set up your clinical trial in such a way that you can test that hypothesis and really speed up the time that it takes to get medicines to the right patient population. An important part of that is that you’re not treating patients when you have data that says they won’t respond.”
There is also great potential for these technologies to be applied at the other end of the clinical cycle, informing the delivery of treatment to patients in more efficient and targeted ways. That is exactly what has been achieved at the National University of Singapore, where a team of researchers have developed the CURATE.AI platform, an AI-driven system which utilises a patient’s data – such as tumour size or biomarker levels in the blood – to identify optimal dosages of therapies for the best possible outcomes.
Professor Dean Ho, Director at the Singapore Institute for Neurotechnology (SINAPSE), National University of Singapore, explained how the technology operates:
“CURATE.AI is [not] an algorithm that models drug synergy, pharmacokinetics, or disease mechanisms,” he began. “Instead, it is a correlation that directly relates drugs and their respective doses – inputs – with quantifiable measures of efficacy and safety – outputs. Some examples would include tumour burden, viral/bacterial load, and liver/kidney panels.
“The platform has shown that the relationship between drug/dose inputs and efficacy/safety outputs can be represented by a phenotypic response surface. Using only a patient’s own data, a surface is constructed from which the optimal dosing can be identified.
“As such only a patient’s own data is used to optimise their care during the entire course of treatment,” he continues. “Importantly, this surface implicitly incorporates everything ranging from disease mechanism to pharmacokinetics, since efficacy and safety are a summation of everything that occurs prior to the treatment outcome. Furthermore, how patients respond to treatment can evolve over time. Therefore, we have shown that dose modulation, or dynamic dosing, can be mediated by CURATE.AI to continuously optimise care for the full duration of therapy.”
As Professor Ho mentions, the platform addresses a crucial factor in the pursuit of continuous, effective treatment – that patient response is not always constant throughout. This is something Dr Rossanese also identifies as a key challenge, and because of this, there is much more to the application of precision approaches than may be first apparent.
Just giving someone a drug and saying ‘this is personalised medicine and we’re done’ probably isn’t right either; we should also include monitoring the specific response of that patient to the drugs that they’ve been given, because we know there is huge variability in the response, and that’s related to the heterogeneity of humans, but also the heterogeneity that we see in cancer,” she notes. “We need to see when a patient is responding to the precision medicine we’ve given them.
“One of the things we’ve identified as a problem is that initially a lot of people respond to therapy, and things look good and they go into remission, but of course they later relapse. What we know now is that is essentially a Darwinian process in which the cancer is evolving to be able to deal with the therapy. How this links to personalised medicine is that we should definitely consider monitoring someone over time, because we know their tumour is going to change.”
Professor Ho also notes this same point as crucial, detailing how CURATE.AI factors this into its method to ensure the highest level of flexibility and efficacy in the delivery of personalised treatment.
“The beauty of CURATE.AI is that it is disease indication-agnostic. Therefore, we don’t build mathematical models to recapitulate different diseases because patients are too different from one another, and the same patient changes during the course of treatment,” he explains. “CURATE.AI can dynamically account for these changes to modulate treatment to maintain optimal responses. The CURATE.AI platform can be applied to virtually all disease indications, as well as there is a quantifiable measure of efficacy and safety.
“On a clinical level, CURATE.AI has been successfully validated in prospective studies for oncology, transplant immunosuppression, and multiple infectious diseases. Multiple trials are ongoing and cleared to start for additional solid and haematologic cancers. In the preclinical space, this technology has been applied to over 40 disease indications with 0 misses. For indications where there isn’t yet an existing marker from which a clinician can quantify efficacy, imaging can be used as well through a properly designed set of serial imaging sessions. We have obtained clearances to start trials based on this output for CURATE.AI as well.”
While these approaches bring remarkable potential to the table when it comes to achieving better outcomes for patients through more informed and targeted approaches, they do not present these benefits without their fair share of challenges. While the use of AI and machine learning can indeed process data on a scale otherwise unattainable, this also provides insight into how we look at other hurdles to be overcome.
“If we say that we need to match the patient and their genetics with the molecularly-targeted drug, you can look at it from two ways,” Dr Rossanese remarks. ”One is that we don’t have enough drugs, so our ability to measure or detect mutations in cancer or molecular characteristics of tumours has actually outpaced our drug discovery in order to have medicines for each of those mutations or characteristics. There’s a mutation in a gene called K-Raf and it’s present in a lot of lung and colon and pancreatic tumours. We can measure that mutation, so we can sequence tumours and we can tell patients they have a K-Raf mutation. We know that a K-Raf mutation is a poor predictor of outcome, and we know that it causes resistance to other targeted therapies – there are drugs that we won’t give you if you have a K-Raf mutation because we already know that that mutation is a predictor of resistance. But what we don’t have is a drug that works against K-Raf to offer those patients. To overcome that particular problem, we need more therapeutics, with different modes of action – it’s basically expanding our molecular toolbox so that we have more medicines to match to more patients.
“The flipside of that is that we sometimes have good drugs that work in a subset of patients, but we don’t know the molecular or genetic predictors of that response. A really good example of that is the experience we’re seeing with immunotherapy; we’re seeing really great responses in a subset of patients, but we don’t’ always know why that is or what defines those patients’ tumours, or even those patients’ innate immune system which is leading to those great responses. On that side, it’s a better understanding of the pharmacology and the molecular mechanism of the drug to know why these patients are responding – then you go find those patients and treat them.”
So how does the importance of technology – particularly AI and machine learning – fit into the outlook of precision medicine in the coming years and beyond? Professor Ho offers some predictions:
“AI will play a key role in bringing precision – using genomic profiles to select drugs – and personalised medicine – individualising drug dosing to a single patient – into widespread use,” he said. “AI will accelerate all of the above capabilities, and make both precision and personalised medicine more actionable so that clinicians will be able to deploy tailored regimens at doses that will continue to mediate effective care for the patient.
“There is a big drive to integrate wearables with mobile/digital applications so that users, both healthy individuals and patients, can streamline data collection and intervention to address disorders,” he added. “Mobile applications and wearables will enable more point-of-care treatment and play a role in democratising medicine.”
Dr Rossanese, too, shared her final thoughts on the matter: “We know a lot more, and so it’s a shame to not use that knowledge,” she remarked. “Because we can define these things, it’s really important to translate them out of the lab, where I know them, into clinical practice. Again, I think it’s about getting to the patients who we think will respond, and more importantly, not doing unnecessary treatments on the patients who won’t, and getting them into their appropriate clinical trials quickly as well. I think it can speed a lot of things up.”
She added: “Everyone always loves to talk about cost of cancer treatment or of drugs in general, but if you’re not treating a lot of people who aren’t going to benefit, then you free some of that money up to treat the people that are, so cost is going to be important as well.”