The unique how:

Active Learning & Differentiated Scaffolds

Addressing these specific problems was a complex chemistry design challenge, one that our combination of generative chemistry and active learning was uniquely positioned to solve. Recursion leveraged its AI-native precision design platform to move beyond the inherent biases of traditional medicinal chemistry. By employing active learning, the platform intentionally sampled unbiased, diverse chemical space to select multiple starting points. These were not only the “best” compounds for optimal predicted CNS penetrance and potency, but also counter-intuitive compounds that deviated from standard design biases - providing alternative starting points that could be optimized.

Through sampling these diverse areas of chemical space and molecules with distinct structural profiles, a novel chemical scaffold was identified. This scaffold was further refined into REC-4539.

By using the platform for unbiased selection of diverse starting points, our models proposed more information-rich initial molecules away from the Pareto front (the mathematical boundary of multi-parameter optimization where one key property e.g. potency cannot be improved without degrading another e.g., safety) and led to a novel scaffold and ultimately REC-4539

The outcome:

A Potential Best-in-Class, Reversible, CNS Penetrant LSD1 Inhibitor

REC-4539 is an orally bioavailable, highly potent, and reversible LSD1 inhibitor designed to overcome the limitations that hindered previous agents and potentially be the first LSD1 inhibitor to deliver on the promise of the target:

• Improved Therapeutic Index: By utilizing a reversible binding mechanism and with a shorter predicted human half-life than competitors’ current reported data, REC-4539 allows for more precise dosing. This would potentially allow for better management of on-target toxicities, thereby reducing the risk of treatment-limiting thrombocytopenia.

• CNS Penetrance: Unlike earlier inhibitors, REC-4539 is specifically designed to cross the blood-brain barrier. This would potentially provide a direct mechanism to reach and impact brain metastases, addressing a primary site of disease progression in SCLC.

• Preclinical Performance Combining Efficacy and Safety Signals: In a preclinical SCLC xenograft model, REC-4539 drove significant, dose-dependent tumor growth inhibition while remaining well-tolerated with minimal impact on platelet counts.

Table showing REC-4539 meets DC criteria for efflux and predicted half-life over competitors.

Compelling dose-dependent tumor growth inhibition demonstrated in vivo for REC-4539 in an SCLC xenograft model, correlated with a decrease in a neuroendocrine tumor biomarker proGRP.