Insilico Medicine announced that its research paper, “When Single Answer Is Not Enough: Rethinking Single-Step Retrosynthesis Benchmarks for LLMs,” has been accepted for presentation at the International Conference on Machine Learning 2026. The study challenges conventional retrosynthesis benchmarking approaches that rely on single “ground-truth” answers and Top-K accuracy metrics, which may not reflect the multi-solution nature of real-world chemistry.

The paper introduces ChemCensor, a chemistry-aware evaluation metric designed to assess model performance based on reaction centers and functional groups, aligning more closely with expert human reasoning. Additional contributions include the CREED dataset, comprising 6.4 million validated reactions; benchmarking results from the C3LM model; and the URSA-expert-2026 dataset, an expert-annotated benchmark designed to reduce data leakage and improve evaluation rigor.

The research supports the development of more realistic and scalable training and evaluation frameworks for AI-driven retrosynthesis and drug discovery. Supporting materials will be made publicly available to promote transparency and reproducibility.

Manganese dioxide can convert amino acids into hydrogen cyanide (HCN) without requiring methane, solving a long-standing puzzle about the origin of this key prebiotic molecule on early Earth, as reported by researchers from Science Tokyo. Although HCN is central to origin-of-life theories, recent evidence suggests early Earth's atmosphere didn’t contain sufficient methane needed for classic HCN-producing reactions. The newly found chemical pathway shows that HCN could instead have been continuously supplied from abundant amino acids.

Magnons are tiny waves in magnetisation and ideal building blocks for hybrid quantum systems and quantum metrology. However, their previously too-short lifetime of at most a few hundred nanoseconds has been a hurdle. An international team of physicists led by Andrii Chumak from the University of Vienna has now succeeded in extending this lifetime a hundredfold to up to 18 microseconds – paving the way for a quantum computer the size of a 1-cent coin. The scientists have also made the crucial discovery that it is not a fundamental law of physics that governs the lifetime of magnons, but rather a question of materials. The study has recently been published in the prestigious journal Science Advances.