Advanced Materials Simulations Group


AMSG AI-Lab (beta)

We develop ML algorithms to accelerate the chemical research. The AI-LAB currently includes user-friendly AI modules for retrosynthesis and reaction prediction of organic molecules, and synthesizability prediction of inorganic crystal structures. More modules with GUI will be updated in due course.

Organic AI



1. Retrosynthesis: LocalRetro

Paper: Deep Retrosynthetic Reaction Prediction using Local Reactivity and Global Attention. JACS Au, 2021
GitHub source


LocalRetro
LocalRetro is an accurate machine learning-based AI for predicting the possible reactans for synthesizing a given moleucle (retrosynthesis) using graph neural networks (GNN) and local reaction tempalte.




2. Reaction prediction: LocalTransform

Paper: A generalized-template-based graph neural network for accurate organic reactivity prediction, Nat Mach Intell, 2022
GitHub source


LocalTransform
LocalTransform is an accurate machine learning-based AI for predicting possible outcomes of organic chemical reactions using graph neural networks (GNN) and generalized reaction tempalte.


Inorganic AI



1. Synthesizability prediction: PU_CGCNN

Paper: Structure-based Synthesizability Prediction of Crystals using Partially Supervised Learning. J Am Chem Soc, 2020
GitHub source


PU_CGCNN
PU-CGCNN is a python code for predicting CLscore (crystal-likeness score) which is quantitative synthesizability metric of inorganic crystals. This is a partially supervised machine learning protocol (PU-learning) using CGCNN classifier (by T. Xie et al.).



The use of our AI-Lab services and platform is regulated by Terms of use
The results of any AI model can be downloaded and used by each user under CC-BY license