Differential attention transformer-enhanced graph neural network for accurate material property prediction

We are pleased to announce our latest journal publication titled “Differential attention transformer-enhanced graph neural network for accurate material property prediction”, published in Engineering Applications of Artificial Intelligence.

This paper introduces a Differential Attention Transformer-enhanced Graph Neural Network (DAT-GNN) for accurate material property prediction. By combining chemically informed structural features with differential attention, self-attention, residual graph convolutions, and positional encodings, the proposed framework effectively models complex atomic interactions in crystalline materials. Experimental evaluations on benchmark Materials Project datasets demonstrate improved prediction accuracy for both formation energy and bandgap, highlighting the potential of DAT-GNN for accelerating materials discovery and advancing data-driven materials informatics.