NISPO: Open-source IUPAC name generation tool

2026-07-28

AI Summary

Existing open-source tools struggle to generate IUPAC-compliant chemical names with high accuracy, hindering standardized representation of chemical structures in scientific literature and databases. This work proposes NISPO, an open-source Python tool built on RDKit that introduces, for the first time, an intelligent agent framework powered by GPT-5.5 and Codex, which iteratively refines its naming algorithm through self-optimization guided by back-translation feedback from OPSIN. By leveraging large-scale training and validation data from SureChEMBL and PubChem, NISPO achieves a round-trip accuracy of 98.1% on a test set of 103 million molecules, substantially outperforming current state-of-the-art methods. The source code has been publicly released to facilitate community adoption and further development.

Abstract

Systematic International Union of Pure and Applied Chemistry (IUPAC) names are standard for communicating molecular structures in chemical literature, patents, and databases. We introduce NISPO, an open-source RDKit-based Python package for IUPAC name generation. NISPO was developed by an agentic self-improvement loop using OpenAI's Codex with the GPT-5.5 model. A generated name was considered correct if the open-source OPSIN tool parsed it back to the input structure. Guided by this objective, the agent implemented and refined NISPO against 2.68 million molecules from SureChEMBL, resulting in the tool achieving 98.1% round-trip accuracy on a held-out set of 103 million PubChem molecules. NISPO is freely available at https://github.com/oxpig/nispo.

Problem

• IUPAC name generation
• molecular structure representation
• chemical nomenclature
• round-trip accuracy
• systematic naming

Innovation

• IUPAC name generation
• agentic self-improvement
• round-trip accuracy
• RDKit-based tool
• large-scale molecular naming