Unlearn

In the rapidly evolving landscape of digital health innovation, Unlearn stands at the forefront of a revolution that blends artificial intelligence with clinical research methodology. By developing AI-powered digital twins capable of simulating patient populations and predicting health outcomes, this forward-thinking company is transforming how we design and conduct clinical trials—particularly in the challenging field of psychiatric disorders. Through rigorous technological advancement and strategic business development, Unlearn has created tools that not only enhance trial efficiency but also provide deeper insights into complex medical conditions. As regulatory bodies worldwide recognize the potential of these innovations, the company continues to push the boundaries of what's possible in medical AI, aiming to integrate its breakthrough technology more deeply into clinical workflows. This case study explores how Unlearn's AI-driven approach is redefining clinical research, offering both practical applications and theoretical advancements for the future of healthcare innovation.

Company Founding and Core Technology

The company's founding in 2017 marked the beginning of a journey to transform generative AI through the lens of real-world systems. Charles K. Fisher and his team—Aaron and Jon, former colleagues from a virtual reality company—saw an opportunity where few real applications had emerged, despite the excitement surrounding generative models in research, particularly through Generative Adversarial Networks (GANs).

Their breakthrough came from revisiting Restricted Boltzmann Machines (RBMs), which Fisher found exceptionally suited for real-world applications. The team appreciated RBMs' ability to introduce controlled noise in the generation process and their undirected nature, allowing effective integration over missing data—a crucial advancement in modeling complex, noisy systems. By merging modern deep learning techniques like GPU training and adaptive optimizers with RBMs, they created what they initially called Boltzmann Encoded Adversarial Machines (BEAMs), now known as BoltGAN Machines (BM).

The company's early strategy involved seeking venture capital to validate the potential of generative modeling, though they admitted lacking a formal business plan at the time. The initial business pitch centered on using generative AI for system simulators across various domains, including health outcomes for clinical trials, gene expression effects, and weather simulation. These simulations would serve as predictive tools for complex, real-world systems, complementing existing deterministic models.

Graham joined the team as the business lead, facilitating the development of a practical application. The focus quickly shifted to simulating longitudinal health outcomes, a domain where their generative models demonstrated particular promise. While they pursued weather simulation as an ambitious goal, the team's expertise ultimately converged on creating digital twins of patient populations. These digital twins would forecast health changes over time, enabling more accurate and efficient clinical trials.

This technology development process set the stage for Unlearn's most significant contributions to clinical trials, as the company continued to refine its approach to digital twin generation and application.

Business Development and Application Focus

The team validated their technology through practical applications across various domains, starting with weather simulation and health outcomes before settling on patient population simulation. Their initial approach involved seeking venture capital while developing a lean startup strategy to test different applications with potential customers.

One of their first successful validations came in the design and implementation of smaller clinical trials that maintained statistical power or achieved greater efficiency without additional participants. This capability, which aligns with guidance from both the EMA and FDA, has been particularly valuable in accelerating Multiple Ascending Dose (MAD) studies for determining safety, tolerability, and optimal dosing.

In parallel with their foundational work, the team recognized the significant challenges in recruiting participants for psychiatric disorders trials. Addressing this, they developed two Digital Twin Generators (DTGs) that represent major advancements in precision medicine. The Schizophrenia DTG was trained on data from over a thousand patients, while the MDD DTG utilized data from more than five thousand patients across multiple trials.

These DTGs have demonstrated substantial potential in transforming psychiatric disorder treatment landscapes. For instance, the MDD DTG has shown the ability to dramatically reduce variance in outcomes measurement across common trial durations, while the Schizophrenia DTG has demonstrated the largest expected variance reduction within six weeks of an acute event.

The platform's predictive insights dashboard offers real-time participant progression monitoring and highlights sensitive clinical outcomes for improved signal detection. This capability allows trial teams to make more informed decisions at interim and end-of-study analysis points while obtaining precise estimates of treatment effects for all measured clinical variables, including outcomes, biomarkers, laboratory results, and vital signs.

Clinical Trial Applications

The Unlearn Platform applies its digital twin technology across three primary areas: accelerating trial timelines, enhancing decision-making through predictive insights, and identifying sensitive clinical outcomes. By enabling smaller, more efficient studies while maintaining statistical power, the platform supports Multiple Ascending Dose (MAD) studies that accelerate safety, tolerability, and optimal dose determination. Single-arm studies can use digital twins as simulated control groups, allowing quicker progression to later trial phases or go/no-go decisions.

During trial analysis, the platform provides real-time predictive insights for all measured clinical variables through its dashboard. These insights enable researchers to track participant progression, identify sensitive outcomes for better signal detection, and obtain precise treatment effect estimates with enhanced statistical power for confident interim and end-of-study decisions. The platform supports all measured clinical variables, including outcomes, biomarkers, laboratory results, and vital signs.

In psychiatric disorder trials, the platform has developed two Digital Twin Generators (DTGs). The Schizophrenia DTG, trained on data from over a thousand patients, provides predicted outcomes for Positive and Negative Syndrome Scale Total Score or Clinical Global Impression Severity endpoints between one and three months post-randomization. It supports standard of care involving first or second-generation antipsychotics for adjunctive therapy trials or active comparisons against other antipsychotics, demonstrating the largest expected variance reduction within six weeks of an acute event.

The MDD DTG targets Hamilton Depression Total Score as the primary endpoint in acute trials between one and three months after treatment initiation. With training data from over five thousand patients across five trials, it enables antidepressant efficacy evaluation for standard-of-care treatments including selective serotonin reuptake inhibitors and serotonin and norepinephrine reuptake inhibitors. Studies have shown it can dramatically reduce variance across common MDD trial durations while supporting responder subpopulation identification and treatment optimization.

Specific Applications in Psychiatry

The Unlearn Platform has introduced two groundbreaking Digital Twin Generators (DTGs) addressing some of psychiatry's most pressing challenges: one for schizophrenia and one for major depressive disorder (MDD).

The Schizophrenia DTG 1.0 has been trained on data from over a thousand patients, with evaluation showing the largest expected variance reduction within six weeks of an acute event. This model targets Positive and Negative Syndrome Scale (PANSS) Total Score or Clinical Global Impression (CGI) Severity endpoints between one and three months post-randomization or post-treatment initiation. It supports standard of care involving first or second-generation antipsychotics for both adjunctive therapy trials and active comparisons against other antipsychotics.

Similarly, the MDD DTG has demonstrated remarkable capabilities, particularly in variance reduction. Trained on data from over five thousand patients across five trials, it targets the Hamilton Depression Total Score (HAM-D) Total Score as the primary endpoint in acute trials between one and three months after treatment initiation. The model has shown the potential to dramatically reduce variance across common MDD trial durations while supporting responder subpopulation identification and treatment optimization.

These DTGs represent a significant advance in precision medicine, offering an alternative and more comprehensive understanding of individual patient responses. In an industry often plagued by recruitment challenges for psychiatric disorder trials, these generators provide a powerful tool for more efficient clinical research.

Industry Impact and Future Directions

The technology has achieved regulatory recognition, earning qualification from the European Medicines Agency (EMA) and alignment with FDA guidelines for trial design. These regulatory approvals have opened doors for broader adoption while establishing a solid foundation for future development.

The platform's impact extends beyond its technical capabilities to reshape clinical trial methodologies. By enabling smaller studies with maintained power or more powerful designs without additional participants, it directly addresses industry challenges and reduces operational costs. This efficiency gain is particularly significant in specialized areas like Multiple Ascending Dose (MAD) studies, where the technology accelerates safety evaluation and optimal dose determination.

In the realm of trial analysis, the platform introduces transformative capabilities through its predictive insights dashboard. This tool provides real-time forecasting of outcomes for all clinical variables, offering researchers unprecedented visibility into participant progression and subpopulation responses. The system's precision extends to enhanced signal detection, particularly in identifying responder subpopulations, which can lead to more effective treatment optimization strategies.

Looking ahead, the company continues to explore uncharted territories in medical AI, driven by their belief in generative models as a cornerstone for future advances. Unlearn remains committed to pushing the boundaries of AI in healthcare, with ongoing research focused on integrating their digital twin technology with existing clinical workflows to maximize its therapeutic potential.