- Python 97.8%
- Shell 2.2%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| .github | ||
| mattervial_perovskites_full_analysis | ||
| modnet_mattervial_matbench | ||
| other_datasets | ||
| results_encoders_and_SI_data | ||
| .gitignore | ||
| README.md | ||
MatterVial Support Data
This repository contains the data and code used to support the findings in the paper:
"Combining feature-based approaches with graph neural networks and symbolic regression for synergistic performance and interpretability"
by Rogério Almeida Gouvêa, Pierre-Paul De Breuck, Tatiane Pretto, Gian-Marco Rignanese, Marcos José Leite Santos
Abstract
This study introduces MatterVial, an innovative hybrid framework for feature-based machine learning in materials science. MatterVial expands the feature space by integrating latent representations from a diverse suite of pretrained graph neural network (GNN) models—including structure-based (MEGNet), composition-based (ROOST), and equivariant (ORB) graph networks—with computationally efficient, GNN-approximated descriptors and novel features from symbolic regression. Our approach combines the chemical transparency of traditional feature-based models with the predictive power of deep learning architectures. When augmenting the feature-based model MODNet on Matbench tasks, this method yields significant error reductions and elevates its performance to be competitive with, and in several cases superior to, state-of-the-art end-to-end GNNs, with accuracy increases exceeding 40% for multiple tasks. An integrated interpretability module, employing surrogate models and symbolic regression, decodes the latent GNN-derived descriptors into explicit, physically meaningful formulas. This unified framework advances materials informatics by providing a high-performance, transparent tool that aligns with the principles of explainable AI, paving the way for more targeted and autonomous materials discovery.
Repository Contents
This repository includes:
-
Results for the autoencoder and previous version of MatterVial (pGNN): The detailed results testing the different descriptor-oriented featurizers (l-MM and l-OFM)and all results obtained from the previously less general package which are now on the SI of the paper.
results_encoders_and_SI_data/
-
Featurized datasets and models: Featurized datasets and final models for the main results of the paper. These are downloaded from FigShare link, a python script is provided to help with this.
-
download_datasets_and_models.py
-
Matbench v0.1 evaluation of MODNet@MatterVial models: All general results that are shown in Table 1 of the paper are shown here including output files and fold by fold metrics.
modnet_mattervial_matbench_results/Correspond to Table 1 of the paper:
Results table 2, focusing on matbench_perovskites dataset:
- MODNet@MatterVial with adjacent models:
mattervial_perovskites_full_analysis
Getting Started
To replicate the results presented in the paper or to apply these methods to your own dataset, check MatterVial repository enhanced with the GNN featurizers, and check the scripts provided along with the results.
Contact and Contributions
For any clarifications, please contact the corresponding authors:
- Rogério Almeida Gouvêa rogeriog.em@gmail.com
- Gian-Marco Rignanese: gian-marco.rignanese@uclouvain.be