ALIGNN-BERT-TL codebase

This repository contains the code for Hybrid-LLM-GNN transfer learning framework to predict materials properties using concatenated embeddings extracted from ALIGNN and BERT models. The code provides the following functions:Generate string representations of provided crystal samples using NLP tools: Robocystallographer and ChemNLPUse a pre-trained BERT/MatBERT langugae model to extract context-aware word embeddings from text descriptions of a given crystal structureConcatenate LLM embeddings with ALIGNN embeddings obtained from feature-extraction-based ALIGNNTLPost-analysis script for model performance analysis and text-based model explanation<br>