Chemcrow

An AI agent designed to accomplish tasks across organic synthesis, drug discovery, and materials design using GPT-4.

Follows a 'Thought, Action, Action Input, Observation' format to iteratively reason, plan, and execute tasks.

Allows the LLM to write and execute Python code directly for complex computations, data analysis, and AI model training.

Accesses relevant and up-to-date scientific information from search engines to expand the LLM's knowledge base.

Extracts pertinent information from scientific documents (PDFs, HTML) to provide precise, well-supported answers with citations.

Modifies molecules using retro and forward synthesis rules to discover similar structures or generate new compounds.

Identifies and provides an overview of functional groups within a molecule to aid in experimental design and compound synthesis.

Converts molecule names or CAS numbers to SMILES representations and vice versa for molecular analysis.

Determines and compares the similarity between molecules and computes molecular weight from SMILES strings.

Automatically checks molecule CAS numbers against lists of chemical weapons and precursors, halting processes if identified.

Identifies explosive molecules from databases like PubChem, issuing warnings or errors during synthesis requests.

Determines whether a particular molecule has been patented, providing compliance information.

Predicts chemical reactions and retrosynthesis pathways based on provided reactants, generating anticipated products.

Devises multi-step synthetic routes for target molecules, including conditions, additives, and solvents, using advanced algorithms.

Recognizes and categorizes specific chemical reactions from its internal database, aiding in understanding mechanisms and conditions.