然而，现有方法面临两大瓶颈：一是多数生成模型仅能处理连续型数据（如图像像素），难以应对分子设计中离散的原子类型与连续坐标共存的多模态数据；二是针对特定性质的定向生成通常需要额外训练模型，成本高昂。

"AI is everywhere and developing new ways to use ... "This project, which will be their final degree project (TFG), actively involves students from the Complutense University in all phases as ...