Termination of the development of blood cancer drug by Schrödinger – Lessons to be Learned from Clinical Trial Setbacks

Schrödinger recently announced the termination of its blood cancer drug program after two patient deaths during a Phase I clinical trial. The drug in question, SGR-2921, is an inhibitor of CDC7, a protein kinase essential for DNA replication initiation and a central player in the DNA damage response (DDR) pathway.

The story behind SGR-2921

Schrödinger’s drug was the product of an innovative approach: the team crowdsourced compound design via its LiveDesign platform, leveraging collective intelligence from 14 designers and next-generation computational modelling. The candidate selected for clinical development, SGR-2921, exhibited very high binding potency (~10 pM) against CDC7. This compound was tailored to compete with ATP in the high cellular ATP environment and was optimized for selectivity and pharmacokinetic properties using next generation modelling techniques, including a Physics-based model for passive membrane permeability.

Despite early signs of monotherapy activity, SGR-2921 was deemed to have contributed to the deaths of two AML patients in the trial, prompting a full halt to its clinical development. Schrödinger attributes the toxicity observed not to compound design flaws but to the CDC7 target itself. Most CDC7-targeted candidates in the industry have been terminated in early clinical stages, suggesting an inherent risk in modulating this pathway. Additionally, SGR-2921 displayed a narrow therapeutic window, and the patients affected were immunocompromised, raising vulnerability to adverse outcomes

Limitations of AI-based Drug Discovery

The SGR-2921 episode highlights both the promise and pitfalls of AI-based approaches in drug discovery. Current AI-driven efforts largely focus on molecular design parameters, such as:

Binding potency, Selectivity and optimization of ADME/PK properties

While these are essential, they represent only part of the drug discovery equation. Although AI excels at molecule optimization, it does not yet fully address challenges like:

• Target validation - ensuring the biology itself is safe and therapeutically relevant

• Preclinical-to-clinical translatability - bridging the gap between animal models and human patients

• Therapeutic window assessment - predicting safe dosing margins in complex biological systems

In the rush to shorten discovery timelines and deliver novel molecules, critical clinical determinants may be overlooked, leading to costly and tragic outcomes, as seen with SGR-2921. This episode is a reminder: Cutting-edge technology must be balanced with rigorous biology and safety assessment.

https://www.bioprompt.in

https://www.schrodinger.com/life-science/learn/case-studies/design-novel-potent-cdc7-inhibitor-development-candidate-high-ligand-efficiency/

..