SCH: Graph-String Transformer and Reinforcement Learning for Design of Cancer Theranostics

This research aims to develop and validate an innovative artificial intelligence (AI)-driven approach for accelerating the discovery of dual-targeted cancer theranostic agents to overcome therapy resistance. The investigation focuses on developing bispecific small molecule inhibitory conjugates (BsSMICs) targeting both stearoyl-CoA desaturase-1 (SCD-1) and fatty acid desaturase 2 (FADS2), two critical enzymes in cancer cell fatty acid metabolism. The specific aims are to: (1) develop an AI algorithm integrating graph-string transformers and reinforcement learning to generate synthesizable molecules for highly specific target-based cancer theranostics, and (2) implement the AI algorithm to generate novel FADS2 inhibitors and construct BsSMICs for validation through in silico, in vitro, in vivo, and ex vivo evaluations. The research methodology combines advanced AI techniques with experimental validation, utilizing a dual-representation approach that leverages both graph structures and SMILES strings to capture molecular properties. The AI system incorporates domain knowledge and interpretability techniques to enhance reliability and transparency. The experimental validation includes synthesis and radiolabeling of promising candidates, followed by comprehensive biological evaluation using established cell lines and mouse models. Success metrics include achieving binding affinities with pIC50 values greater than 8 for both SCD-1 and FADS2 inhibition and tumor-to-muscle ratios exceeding 3 in imaging studies. This research represents a significant advancement in theranostic agent development, potentially offering new strategies for addressing cancer therapy resistance through targeted molecular imaging and therapy. RELEVANCE (See instructions): This research addresses a critical need in cancer treatment by developing AI-driven methods for novel dual-targeted cancer theranostics that can overcome therapy resistance. These bispecific molecular agents will provide clinicians with the capability to conduct image-guided patient selection for effective treatment of cancer patients.