Predicts a Crystal-Likeness Score (CLscore) for inorganic crystal structures using positive–unlabeled learning with bagged crystal graph convolutional neural networks (PU-CGCNN).
CLscore ∈ [0, 1] estimates how “crystal-like” / synthesizable a structure is relative to experimentally reported materials. It is complementary to thermodynamic filters such as energy above hull — high CLscore does not guarantee experimental success, but is a useful soft prior for screening and generative filtering.
Model: Jang et al., JACS 2020 (pretrained 100-bag ensemble). Paper: https://doi.org/10.1021/jacs.0c07384 Code: https://github.com/kaist-amsg/Synthesizability-PU-CGCNN
Usage
26 callsPU-CGCNN validation: experimental crystals vs high-e_hull theoreticals
Spot-check of PU-CGCNN on 10 experimentally known vs 10 high-e_hull theoretical Materials Project structures. Directional separation, real overlap, and some surprising misses (LiFePO₄, AgC₂N₃).
Will it actually form? Predicting crystal synthesizability with PU-CGCNN
An explainer for the new Crystal-Likeness Score (PU-CGCNN) service: what the CLscore means, how positive–unlabeled learning works, and when to use it in a screening pipeline.
Start here: magnet discovery on Ouro
A guide for new researchers: the magnet-relevant services on Ouro, what each is good and bad at (including on rare-earth compounds), how long it takes, and how to tier your search so DFT only runs on compounds that earned it.
Everyone is a scientist. Science needs compute. Why Ouro wants to build with Modal.
From Ouro's founder: why the platform's public science services run on Modal, what's already live, and an open invitation to partner so open science tools can stay free and anyone can publish their own.