In Silico Discovery of Novel KRAS G12D Inhibitor Candidates

Defensive publication of novel chemical scaffolds targeting the KRAS G12D mutation (PDB: 5Y0Z).  <-- please read the edit at the bottom, mistakes happend 

Primary Candidate SMILES: N#CC1(CS(=O)(=O)N2Cc3ccc(Cl)cc3[C@H](C(=O)Nc3cncc4c(C=O)cccc34)C2)CC1

Methodology: Hybrid Evolutionary De Novo Design Engine (Atom-Based) combining Vina docking and deep-learning filtration. JarvisAutoSearh (AGI like Agent ARES with Virtual Lab)

Results:- Predicted Affinity: -5.5 kcal/mol- Ligand Efficiency: -0.16- Novelty: Structure absent from public databases as of Jan 2026.

This dataset is released into the Public Domain to establish Prior Art and prevent patenting by third parties.Disclaimer :The molecular structures (SMILES) provided in this report may appear exotic or non-academic compared to standard organic databases. This is a deliberate architectural choice of our AGI-driven pipeline, which is designed to explore the full periodic table beyond restricted organic chemistry.Universal Chemistry Route: To prevent system crashes while handling transition metals (Gold, Platinum) or complex metalloids, the current engine uses a "Shape Proxy" methodology.Intended Behavior: These non-standard notations are not bugs; they are functional placeholders that allow the AGI to maintain physical stability during high-throughput docking and evolution cycles.

Transparency First: This initial release prioritizes transparency. We choose to showcase the model's current strengths (multi-domain exploration) and known weaknesses (SMILES academic formatting) rather than filtering out groundbreaking organometallic results.

Future Roadmap: We are actively developing a refined "Chemical Translator" for upcoming versions to align these results with standard IUPAC/academic SMILES without losing the physical data of the metal centers.

I apologize for any friction these non-standard strings may cause during analysis. We are pioneering a path where AI logic meets physical reality, and your feedback is essential for this evolution.

Contribute to further research & development: > https://ko-fi.com/jarvisautosearch_ares

Edit 1/30/2026 : A refined/filtered dataset will follow. This update will remove noise (fragments/unstable structures) using Jarvis V6's "Scientific Validity Judge" and "Synthesis Difficulty" filters to provide high-confidence drug-like leads . On a + 4000 entry raw datasetEdit 1/31/2026 :  Important correction regarding the target protein. Due to a pipeline configuration error, the previous dataset was generated using the 5Y0Z (SIRT2) PDB structure instead of the intended KRAS G12D model. I apologize for this targeting oversight; the SIRT2 structure was not properly cleared from the Jarvis-ARES docking environment prior to the run.

As a result, the affinity scores in Version 1 reflect binding to SIRT2. A corrected dataset (Version 2) specifically targeting the 7RPZ (KRAS G12D) structure is currently being processed by JarvisAutoSearch via ARES_II and will be uploaded shortly. Thank you for your understanding as we refine this autonomous discovery path.