dotcraft-condgraphvae: Initial Publication Release (v1.0.0)

Abstract

DotCraft: Conditional Graph VAE for Nitrogen-Doped Graphene Quantum Dot Generation Overview DotCraft is a computational framework for the conditional generation and structural refinement of nitrogen-doped graphene quantum dots (N-GQDs). This repository provides the publication-associated model implementation for the study entitled "AI-Optimized Nitrogen-Doped Graphene Quantum Dot Ameliorates Metabolic Dysfunction-Associated Steatohepatitis by Targeting ACC1." The framework integrates a conditional graph variational autoencoder (Conditional Graph VAE), rule-constrained graph construction, and structural validation procedures to support the computational design of graphene quantum dot candidates. Model Architecture and Functionality The Conditional Graph VAE acts as a graph-profile refinement model rather than a fully unconstrained atom-by-atom molecular generator. It predicts structural descriptors, including heavy-atom counts, elemental composition, mean graph degree, bond-type fractions, and condition-reconstruction outputs. Predicted structural profiles are subsequently realized through fixed-core or ring-growth graph construction strategies. Generated candidates undergo structural validation using predefined topology and geometry constraints to ensure consistency with the specified design conditions. Dataset and Reproducibility The v1.0.0 release includes: 16 canonical graphene quantum dot parent cores and 176 fixed-core structural classes. A predefined design space comprising 4,640 condition states. A curated dataset of 297 graph-coordinate pairs, partitioned into training (203), validation (41), testing (36), and holdout (17) subsets. Source code for model training, evaluation, and rule-constrained candidate generation. A released model checkpoint, configuration files, and dependency specifications. Reproducibility utilities, automated tests, and structural validation procedures. Experimental definitions supporting multi-seed training, held-out evaluations, and component ablation analyses. The repository supports independent validation of the released model components and reproducible evaluation of the Conditional Graph VAE on the provided dataset. Scope and Limitations This release represents the computational model component of the broader research workflow. The large-scale production library comprising 396,841 unique retained structures is not included in the public repository. Its integrity can be verified using the provided validation tools when the corresponding external data are supplied. Downstream candidate screening, ADMET assessment, quantum chemical calculations, molecular dynamics simulations, ACC1-related investigations, and experimental biological validation are outside the scope of this software release. Consequently, this release should not be interpreted as a complete reproduction package for all computational and experimental findings reported in the associated manuscript. Software Availability The source code, model configurations, released checkpoint, evaluation scripts, and reproducibility documentation are publicly available on GitHub. GitHub repository:https://github.com/zhangjinquan-group/dotcraft-condgraphvae GitHub release (v1.0.0):https://github.com/zhangjinquan-group/dotcraft-condgraphvae/releases/tag/v1.0.0 License This software is distributed under the MIT License.