Conformation Generation

Molecular conformation generation aims to predict the three-dimensional structure of molecules, a crucial task in drug discovery and materials science. Recent research heavily utilizes deep generative models, particularly diffusion models and graph neural networks, often incorporating physical constraints like force fields or operating in SE(3)-invariant spaces to improve accuracy and efficiency. These advancements address limitations of traditional physics-based methods by generating diverse and accurate conformations faster, impacting areas like virtual screening and property prediction. A key focus is mitigating biases in training and improving sampling diversity and speed.