Intrepid Labs’ self-driving lab reimagines drug formulation

The pharma industry has long relied on tried-and-true drug formulation approaches sometimes rooted in trial and error, but this reliance on the familiar has created a bottleneck in the development of new and better medicines. University of Toronto Professor Christine Allen, Ph.D., co-founder and CEO of Toronto-based Intrepid Labs, points out that the traditional drug formulation approach severely restricts the white space of formulation design space: “If you go through the formulation literature, you’ll see hotspots,” she said. “The vast majority of the formulation work is in one small little corner.

This reliance on past successes creates a self-limiting cycle. “There’s a lot more exploiting what’s worked than being really creative and innovative and trying something that nobody’s ever tried,” Allen added. “That’s because we all go back to what’s worked for others.”

A similar challenge extends across the pharma landscape: the data used to train traditional models is often incomplete and biased given the considerable amount of proprietary industry data and the tendency to focus on successful outcomes. “Most of us don’t publish our negative data,” Allen said. A 2024 analysis in Nature’s Scientific Data noted that selective reporting and publication bias remain common in medical journals. “I’ve got filing cabinets of negative data from 20 years as an academic, but journals usually don’t want to see that.” This creates a skewed perspective, where AI models are trained primarily on successes, leaving them ill-equipped to navigate the full complexity of drug formulation.

Self-driving labs can travel across vast formulation spaces

In the case of Intrepid Labs, self-driving labs tap ML models to propose new drug formulation designs based on previous experimental data. Robotic systems executing the proposed experiments autonomously in the company’s lab, and then feed results back into the machine learning models to further refine and improve the formulation designs in a closed-loop process. “Unlike the approach of collecting a lot of data or curating large datasets, we can start with no data,” Allen said. “That’s really one of our greatest strengths.”

Intrepid Labs’ ML-driven platform is capable of exploring expansive formulation design spaces, sometimes encompassing as many as 1 billion possible formulations, to identify novel, optimal formulations. The company aims to accelerate therapeutics development while optimizing properties and performance via a self-driving lab that integrates advanced machine learning and robotics. In essence, a self-driving lab is a “machine-learning-assisted modular experimental platform that iteratively operates a series of experiments selected by the machine learning algorithm to achieve a user-defined objective,” as Nature noted in 2023.

Breaking down how it works

Intrepid Labs’ website notes the company’s mission is to “fast-track therapeutics development with improved properties and performance.”

Here’s how it works in the real-world: A client provides Intrepid Labs with a target product profile — say, an oral formulation of a hydrophobic drug with specific release characteristics. Intrepid inputs this information, along with the drug itself and parameters for the formulation design space (e.g., whether it should be a capsule or tablet), into its proprietary algorithm.

The algorithm then goes to work, selecting which formulation to prepare and characterize first. This initial formulation is created on a small scale using automated workflows within Intrepid’s self-driving lab housed at JLABS, a global network of life science incubators run by Johnson & Johnson Innovation. The results — information about the formulation’s properties and how well it meets the target profile — are immediately fed back into the algorithm. “We take all of that data, it’s fed into the model, which is then retrained and identifies which formulation to prepare next,” Allen said. “We go through this loop several times, generating a series of batches, each one informed by the results of the one before it.”

Humans in the loop

While the automated system handles the rapid-fire experimentation, humans ensures the process stays on track and leads to meaningful outcomes. Intrepid Labs’ team of scientists, with decades of combined experience in drug formulation, keep an eye on the output of every stage.

For her part, Allen has 20 years of experience in drug formulation and development while co-founder and chief AI advisor Alán Aspuru-Guzik is a serial entrepreneur and professor of chemistry, computer science, chemical engineering and materials science at the University of Toronto. Rounding out the co-founding team are chief scientific officer Pauric Bannigan, Ph.D., who has extensive experience in designing drug delivery platforms, and director of research and development Riley Hickman, Ph.D., an expert in applied mathematics, software development, machine learning and robotics.

Real-world results

This collaborative approach, where human intuition and experience work in tandem with the speed and analytical prowess of a bespoke ML system, is already yielding results. In a recent project, a company approached Intrepid Labs with a challenge to create a capsule formulation of a drug with an extended-release profile. “We told the algorithm this is the drug we need to work with and we gave it all possible ingredients that needed to be included,” Allen said. “And we told it that it needs to achieve a nice linear release profile.”

Through a series of automated experiments, the platform got to work. In just 11 batches — a matter of days — the platform had pinpointed a “right-on-target” formulation that met the desired extended-release profile. Intrepid Labs simultaneously optimized granule geometry, granule size distribution as well as the drug release profile. The design space for the project included 12 polymer options — both single or in combination, various processing parameters, base formulation and dose. In all, the total design space included more than 1.06 billion possible experiments.

A new breed of scientist

Ultimately, Allen sees the company’s synthesis of ML and robotics with scientific acumen as a harbinger of what’s to come. “We’ve got these young people working at the University between my lab and Alán Aspuru-Guzik’s lab” who have expertise in drug formulation and machine learning or robotics. “These people can speak both languages and are right at the interface, making those connections,” she said. “You need those people who can really cross-talk and work together.”

At Intrepid Labs, the idea is to drive collaboration between machine learning engineers and drug formulation researchers. “We mix it up because that’s where the power lies,” Allen said. Intrepid Labs looks for individuals who have expertise in both domains when looking for talent. “I know 10 years from now, there’ll be many of these people, but right now, they’re not easy to find.” When the new generation of tech-savvy scientists emerges, Allen predicts:  “It’s going to be phenomenal.”

---

Filed Under: clinical trials, Drug Discovery, Industry 4.0, machine learning and AI
Tagged With: AI-driven drug development, automation, drug formulation, Intrepid Labs, machine learning, optimization, robotics
 

---