Work

XMolCap: Advancing Molecular Captioning Through Multimodal Fusion and Explainable Graph Neural Networks