maranasgroup/MechFind: MechFind v1.0.0

Date: February 17, 2026

Commit: fcc0896

Repository: maranasgroup/MechFind

Initial Release: MechFind v1.0

We are proud to introduce MechFind, a computational framework for the de novo prediction of detailed enzyme reaction mechanisms. MechFind bridges the "mechanism gap" in bioinformatics by generating elementally and charge-balanced mechanistic hypotheses using only overall reaction stoichiometry as input.

This release accompanies our publication: "MechFind: A computational framework for de novo prediction of enzyme mechanisms" (Hartley et al., 2026).

Key Features

• Stoichiometry-Only Input: Predicts mechanisms without requiring 3D protein structures or user-supplied active site residues.

• Moiety-Based Abstraction: Uses a novel graph-based encoding where reaction steps are modeled as the gain or loss of specific chemical moieties (defined by canonical SMILES).

• Hybrid Optimization Strategy:

  • Parsimony: A Mixed-Integer Linear Programming (MILP) formulation (minRules) identifies the fewest number of steps required to balance the reaction.

  • Ordering: A secondary formulation (OrderRules) determines a chemically feasible sequence for those steps.

• Similarity Re-Ranking: Scans candidate mechanisms against the Mechanism and Catalytic Site Atlas (M-CSA) to re-rank predictions based on their resemblance to known, validated biological chemistry.

• High-Throughput Capability: Capable of processing large databases; benchmarked on 14,000+ reactions from the Rhea database.

Included Data

• Curated Rule Set: Includes Unique_Rules.csv, a matrix of 4,091 elementary reaction rules derived from 734 manually curated M-CSA mechanisms.

• Arrow Environments: Includes M-CSA_arrow_rules_r0.json, containing the electronic arrow-pushing environments used for the similarity scoring algorithm.

• Validation Datasets: Pre-processed reaction SMILES for benchmarking against M-CSA and Rhea entries.

Installation & Dependencies

MechFind is written in Python 3.8+ and runs via Jupyter Notebooks for easy interaction.

Dependencies:

• rdkit (Cheminformatics backend)

• pulp (Linear programming interface)

• pandas, numpy (Data manipulation)

Quick Start:

git clone https://github.com/maranasgroup/MechFind.git
cd MechFind
pip install -r requirements.txt
jupyter notebook MechFind_example.ipynb

Usage

The release includes a demo notebook (MechFind_example.ipynb) that walks users through:

1. Loading the elementary rule database.

2. Defining a target reaction (substrate/product SMILES).

3. Running the MechFind prediction pipeline.

4. Visualizing the predicted mechanisms as step-by-step moiety change matrices.

License

This project is licensed for non-profit/non-commercial use only. See the LICENSE file for details regarding commercial licensing via Penn State University.