The Potential of Cell Simulation for Drug Development

AI is transforming drug discovery, but accurately predicting patient responses remains challenging. Considering the ongoing need to connect drugs with the right patients and rising drug costs, scientists seek innovative tools that adhere to fundamental rules of biology.

Turbine has developed a leading interpretable cell simulation platform for better drug development, based on the principle that human cellular behavior is driven by proteins and their interactions. In this Innovation Spotlight, Daniel Veres, chief scientific officer at Turbine, discusses how Turbine’s Simulated Cell™ digital lab excels at predicting cellular responses to perturbations using protein signaling logic for machine learning training of disease models.

What are the key limitations of traditional wet-lab experiments with respect to developing drugs for personalized medicine applications?

With recent shifts in global trends and macropolitical governance of preclinical testing in research and development (R&D), many believe it is time for in silico experiments to replace wet lab testing. However, there is a key difference between computational predictions informing or guiding the testing and validation processes versus replacing them altogether.

We are trying to create the best of both worlds: the scalability, flexibility and efficiency of a rapidly evolving in silico model system and real-life validation of hypotheses for tangible reassurance. The right regimen is challenging and data-intensive to identify. For example, wet lab experiments have a higher-than-average translation rate for oncology, but this isn’t true for other sectors.

Building in silico approaches is sensitive to the quality and number of datasets used during the learning process. As a result, finding more efficient ways to deliver patient-predictive models that translate to the clinic with even scarce verified data is increasingly important.

What are the fundamental criteria for successfully using AI to connect drugs with the right patients?

We need to find the right drugs for the right patients and not the other way around. This trend is already becoming apparent. A market-led R&D strategy is bound to deliver higher impact and has more chance of clearing the later clinical phases than opportunistic strategies that attempt to make whatever is available in a pipeline work.

This means that drug programs need a positioning strategy from the discovery stage that sharpens towards the end of the development but generally evolves in a valid direction based on insights from available data. An approach that thrives in early positioning can help researchers prioritize available targets and ideas, making pipeline-related commitments more tangible and easier holistically.

What are some examples of AI-based solutions that are designed to improve the efficiency of drug development? How do they differ from one another?

Most companies are already using different flavors of machine learning-based approaches for curated or purpose-generated single datasets. The future lies in harmonizing many of these datasets and finding a way to extract foundational logic that can enable delivery of predictive analytics that generalize beyond training data. Foundation models show promise in this regard but need a lot of data to become predictive enough, which is a quality and efficiency limitation.

What is the Simulated Cell and how does it fare compared to wet-lab experimentation?

At Turbine, we try to combine the best of both worlds. Our Simulated Cell approach can generalize and deliver predictive analytics without millions of perturbed patient datasets, which are unfeasible and potentially unethical to attain.

Our virtual lab is designed with a foundation model-like architecture that can be trained on multimodal data to develop a rich feature set. We can make and perturb virtual cells with this technology that can learn how to behave like perturbed patient cells. This model has the capacity to learn from even in vitro verified data to feasibly, reliably, and quickly generate across many modalities, diseases, and samples.

How has Simulated Cell been applied to real-world efforts for improving drug development processes?

Our virtual lab enables computational prediction of therapy effects, allowing drug developers to focus resources and substantially increase the likelihood that new treatments will make it to the patients who need them most. We spent the past decade developing our Simulated Cell technology and building a virtual oncology lab. We aim to spend the next decade delivering this technology to the public and tackling complex diseases beyond oncology, including immune-related or neurodegenerative diseases.

Simulations have been validated through partnerships with leading pharma and biotech companies, including Bayer, AstraZeneca, Ono, and Cancer Research Horizons. Turbine continues to integrate Simulated Cell with other AI-driven discovery tools and contract research organizations worldwide.

What excites you most about the future potential of AI drug development tools like Simulated Cell?

I see a future in which wet lab testing is used as an efficient and controlled data generation step for feeding patient-predictive in silico preclinical screens that “simulate” how patients will react to therapies. Think of it as a computer-aided design for patient biology or a factory simulation before a Formula 1 car hits the asphalt for the first time in a season.