Metabolic reconstruction is a challenging and time-consuming process, often resulting in slow progress in developing new reactions. An open-source and accessible interactive interface that allows users to test their curated reactions quickly and accurately can significantly improve efficiency.

We present Reconstructor, a manual and AI-enabled reconstruction interface that allows researchers to analyze curated reactions and generate predicted reactions. Reconstructor accurately provides mass and charge balance information related to curated reactions, offers detailed information on metabolites, and generates atom mapping of the reaction. Reconstructor features a GPT-enabled prediction function where users can input a gene and receive predicted reactions. This predictive capability has been demonstrated to be somewhat accurate in certain types of metabolism when compared to known reactions in a reliable database.

Reconstructor allows various input types for metabolites from reference databases such as VMH, ChEBI, SwissLipids, and MetaNetX formats. It also includes a drawing function using ChemDoodle, with standardisation performed using RDKit. Reactions are created by verifying metabolites, generating RXN files, and performing atom mapping via Reaction Decoder. Mass and charge balance are checked and detailed molecular information, including formula and atom counts, is provided.

Users can save their curated reactions, name them (and optionally add them to a group) for easy access, and edit them through the interface.

Reconstructor supports integration with metabolic databases like VMH. As a test of this integration, we used a copy of the VMH database and have successfully integrated the interface with it, ensuring smooth operation. The integration process includes validation to avoid duplicates, preparation of reaction data, submission to the database, and confirmation to the user. This integration ensures up-to-date datasets and facilitates efficient data management across platforms.

Availability and Implementation: The interface is available at http://reconstructor.humanmetabolism.org/ and the source code can be found at https://github.com/opencobra/reconstructor.