An AI-guided proteolysis-targeting chimera achieves picomolar degradation potency against a fusion oncoprotein that existing kinase inhibitors cannot fully suppress, according to a paper published August 11 in PNAS. The molecule, DCL05, selectively degrades the entire CLIP1–LTK fusion protein eliminating both its kinase activity and its scaffolding function at a DC50 of 40 picomolar.
The CLIP1–LTK fusion drives tumorigenesis through two cooperating mechanisms: constitutive kinase signaling and CLIP1-mediated multimerization. Current ALK inhibitors target only the catalytic domain, leaving the scaffolding function intact and creating a route for resistance. By integrating deep-learning ternary-complex prediction with structure-based molecular optimization, the researchers designed DCL05 as an orally bioavailable degrader that clears the full fusion protein rather than merely blocking one active site as an approach that aligns with a broader wave of AI-driven drug design transforming oncology pipelines.
In preclinical models, DCL05 outperformed standard kinase inhibitors across a broad panel of LTK resistance-associated mutations, both in vitro and in vivo. The paper also identifies specific resistance-conferring mutations that informed the AI-driven design loop—data the authors say establishes a generalizable pipeline for PROTAC development against other kinase-driven cancers. The ternary-complex predictions that guided DCL05's optimization are part of a growing toolkit: new AI models are also accelerating molecular simulations for drug development, cutting the time needed to map how candidate molecules behave at atomic scale.
The approach addresses a central problem in targeted oncology: tumors that bypass inhibition by exploiting noncatalytic protein functions. PROTACs represent a growing frontier for conditions where conventional small molecules fall short.