Inspiration
The design of clinical trials occurs at the junction of chemistry, pharmacology, variability in patients, and various other challenges of the real world. This led to the inspiration that the teams should explore all of these aspects together rather than using tools and spreadsheets. If scientists were able to change a molecular group, modify how a drug is delivered, or alter virtual populations of clinical trials, then they would be able to see how these changes would impact the clinical aspects of the trial.
What it does
Trialbench is a versatile virtual clinical-trial tool that helps drug development teams test different drug and trial design possibilities. Users can modify parts of a molecule, the route of delivery, dosing, and criteria for including different groups of patients. The tool can show how these changes could affect a sample of patients with respect to drug exposure, efficacy, side effects, and various subgroup parameters. Users can also check the effects of factors such as age, kidney function, diabetes, and metabolism, as well as get how concentration curves and uncertainty vary. The tool does not provide a straightforward prediction, but rather allows users to clarify various assumptions, uncertainties, sparse data populations, and limitations of the model.
How we built it
Trialbench was developed as a complete stack system, comprising a front-end, back-end, a simulation engine, and an API layer. The front-end part facilitates user interaction in the sense that it allows users to manage molecular fragments, drug delivery parameters, dosage approach, and virtual patient requirements. In response to user input, the back-end part executes the simulation without exposing the physical model. The simulation engine integrates simplified molecular properties with pharmacokinetic equations describing absorption, elimination, distribution, and steady-state drug concentration in a human body. It creates a virtual patient population along with various characteristics, including age, weight, kidney function, diabetes, and vitamin status. The run of the simulation lets doctors experiment with the parameter settings and see how these influence the therapy outcomes. The system also performs uncertainty simulation, which modifies the input simulation parameters as well as makes it possible to conduct subgroup analysis, although in this case, the simulated results have external, non-validated nature.
Challenges we ran into
A significant obstacle we faced was how to simulate interactions between chemical, dose, delivery and personal factors in a simple manner. We needed to explain uncertainty in an appropriate way instead of treating simulating as an accurate prediction. Another important issue, users required reproducibilty of simulations even when trying out different designs, and this made us consider using some virtual populations and the simulation of uncertainty in this case. Finally, the migration from initial prototype towards backend, frontend and API involved technical difficulties, which can be seen from the build log that shows that during the first deployment of the system developers encountered an error while creating a new directory.
Accomplishments that we're proud of
We take pride in the fact that we created a unique interactive environment exploring molecular design and clinical trial design simultaneously. Whereas scientists typically have to use different tools, Trialbench links alterations in chemistry and trial design directly to a virtual cohort of patients and indicates the possible impacts such changes might have on exposures, effectiveness, negative outcomes, and subgroups. It is also worth mentioning that we restructured the prototype into multiple modules to separate the simulation engine from the interface, making it easy to test and develop it further.
What we learned
The understanding we gained is that a valuable simulation must not just give a single output, but facilitate the understanding of the assumptions and the uncertainty behind the output. We also understood the importance of connections between molecular properties, pharmacokinetics, dosage, clinical features, and eligibility for clinical trials since any changes within one category might influence the others. From an engineering aspect, separating core simulation engine and interface simplifies the process of rationale, testing, and using in preparation for future integration of backend and API.
What's next for Trialbench
Our forthcoming plan is to enhance the Trialbench prototype into a full-fledged virtual system that entails improving the pharmacokinetic model, where we shall add not only various patients’ profiles but also different patient responses and adaptation estimates. Risks and limitations of the system will also be made more transparent. We will enhance the model further by adding graphical presentation, customizable experimental designs, and APIs for running and analyzing simulations. Ultimately, the Trialbench approach will allow for collaborative experimental design and reporting of reproducible simulations while enabling users to see and assess the assumptions and uncertainties involved in the process.
Log in or sign up for Devpost to join the conversation.