We鈥檝e all seen the statistics. Drug development has a scarily high attrition rate, with approximately 90% of candidates falling short. R&D failure is particularly persistent in the oncology field, where attrition rates are estimated at greater than 95% due to a frequent disparity between a drug鈥檚 positive preclinical results and subsequent inactivity in patients.[i] The costs of oncology trials typically exceed those of other therapeutic areas, and recent estimates suggest that the industry spends approximately $60 billion every year on unsuccessful cancer drugs.[ii]
This is often because researchers don鈥檛 fully understand the cause-and-effect relationship between a drug and its target, which is not surprising given the majority of pre-clinical testing is conducted with isolated systems and animal models that fail to replicate the disease accurately. In addition, the researchers may have targeted unsuitable patients in clinical trials, implemented a bad choice of trial design, or selected the wrong biomarkers or endpoints, resulting in studies that do not meet their objectives despite the molecule鈥檚 potential.
To summarise the issue, the high failure rate in clinical research is a product of the complexity of human biology, with lab experiments often miles away from patient response. This means that, not only are billions of dollars being wasted year on year while patients with high unmet medical needs continue to go without effective treatment, but clinical trial participants are also being exposed to drugs that may carry toxicity without any therapeutic effect.
What about AI?
The critical need to improve the odds of drug development is clear, requiring better in silico technologies for understanding complex disease mechanisms. For the best part of a decade now, there has been considerable hype around the ability of artificial intelligence (AI) to do just this, with AI and big data being rated as the two most impactful technologies in the industry for five consecutive years in GlobalData鈥檚 annual The State of the Biopharmaceutical Industry survey. One idea is that computational models can be used to study and predict the behaviour of complex systems, including the reaction of new molecules with biological targets, thus accelerating the identification of compounds with therapeutic potential.
However, the lack of truly patient-representative data has created pitfalls with this approach, with none of the training-grade datasets close enough to real patients to generate meaningful results. Sometimes, models perform well on paper but fail to predict patient response accurately.iii The first wave of AI-designed drugs has faced significant challenges in the clinic, leading to failures.
鈥淭o shift the paradigm in R&D, we need AI to understand how a novel therapy or target will function where it matters most 鈥� in patients,鈥� says Daniel Veres, chief scientific officer and co-founder of Turbine Simulated Cell Technologies, a Hungarian company that has engineered a better way of using AI for predictive computational models of human cells and tissue.
Introducing Simulated Cell鈩�
Turbine integrates and harmonises multiple public and proprietary OMICS data sources into a universal cell signalling network, constantly extended by an in-house active learning approach to data generation. The team leverages machine learning to train patient-representative avatars that react to a variety of triggers exactly how real cells would,iv thus driving better understanding of the disease and our cells鈥� mechanisms of response and resistance. With avatars comes the option of unlocking computational speed and scale, and with mechanistic understanding comes the capability of matching the right patient cohort to the right therapy or combination and treating even the most complex diseases.
鈥淲e believe that at a deeper level, all cellular behaviour is driven by proteins and their interactions,鈥� says Kristof Szalay, CTO and founder of Turbine. 鈥淥ur platform learns this universal language from scalable experiments and applies its knowledge to predict the response of models it has never seen before, ranging from engineered cell lines to patient samples.鈥�
鈥淲ithin just days and using small datasets typically available in an R&D programme, Turbine fine-tunes this universal model to avatars of biological samples often unavailable for wet lab testing. These avatars are then used in billions of interpretable simulations run at computational speed; a scale impossible to match in real life,鈥� Szalay adds.
As promising hypotheses are ranked and taken forward, Turbine鈥檚 digital lab helps scientists design the optimal preclinical experiments for wet lab validation. In addition, as they seek to translate their hypotheses to the clinic, the platform will help them identify the ideal patient cohorts to benefit from a therapy, aiming to speed up the process and increasing the chance of success.
Not only is this facilitating virtual experiments in volumes that are impossible to achieve in the wet laboratory, but it is also doing so within just days. This robust experimentation can significantly reduce development risk and the price of R&D failure.
鈥淥ur breakthrough technology generates actionable results in half the time that would be required to train other AI models. It also helps explain the biological mechanisms at play, which, on one hand, reveal hypotheses with greater potential to benefit patients and, on the other, identify the exact experiments needed for validation,鈥� explains Szabolcs Nagy, CEO and co-founder.
Moreover, since all its partners鈥� R&D programmes run on the latest version of the platform, the training benefits accumulate, meaning that the technology鈥檚 coverage and understanding of human biology is constantly growing. The future is exciting, but key milestones have already been achieved. Simulations have been validated from target discovery to patient stratification and life cycle management through Turbine鈥檚 collaborations with pharma companies and research institutions, including AstraZeneca, Bayer, Gingko Bioworks, and Cancer Research Horizons.
Of Turbine鈥檚 list of partners, AstraZeneca and Ono Pharmaceuticals are some of the most recent to enroll simulated experiments. Turbine, for example, used its digital lab to find and rank multiple novel targets for further development in one of Ono鈥檚 priority cancer biology domains. Turbine will continue with validating select targets in vitro.
Seishi Katsumata, Ono鈥檚 corporate officer/executive director, said: 鈥淭urbine has provided multiple targets that have first-in-class potential, along with the biology-driven insights necessary to rapidly validate this potential. We鈥檙e excited to move these targets forward and expect that the collaboration will consequently lead to providing a new therapeutic option to cancer patients as soon as possible.鈥�v
With platforms like Simulated Cell, it becomes possible to identify the right drug targets and predict their interactions with greater accuracy and speed, significantly increasing the chances of success in the clinic and reducing the time it takes to get breakthrough therapies to patients.
To learn more about Turbine鈥檚 technology and case studies, please download the deck below.
iii
iv
v
