Fig. 3: LLM-based agents facilitated substrate scope and condition screening.
From: An automatic end-to-end chemical synthesis development platform powered by large language models
a Workflow for substrate scope and condition screening copiloted by Experiment Designer, Hardware Executor, Spectrum Analyzer, and Result Interpreter agents. b The aerobic alcohol oxidation reaction screening task described in natural language for subsequent large language model (LLM)-based agent understanding and OT-2 liquid handler reaction execution. The exact transcript of the natural language description of the task is provided in Supplementary Table 5-7. The image of Opentrons OT-2 liquid handler was obtained from the Opentrons website (www.opentrons.com.cn). c The interaction between human chemists with Spectrum Analyzer for gas chromatography with parallel flame ionization detector and mass spectrometer (GC-FID-MS) result analysis for 3 s to 3p in CuCl2 and DBU condition (see details in Supplementary Table 11). The dagger symbol indicates that the numerical results were generated by the agents’ code interpreter. d The mass spectra for the retention time within 4.243-4.302 min, matched with substrate 3s by Spectrum Analyzer. e The mass spectra for retention time within 3.746-3.800 min, matched with product 3p by Spectrum Analyzer. The red lines represent the characteristic charge ratio signals detected by the Spectrum Analyzer in mass spectrometry, which match the molecule fragments. f Visualization of GC-FID spectra peak integration for substrate 3s and product 3p given by Spectrum Analyzer.
