lingcon01/LiTEN: LiTEN

LiTEN (Learning Interactions with Tensorized Equivariant Networks) is a novel equivariant neural network designed for high-precision, efficient 3D molecular property prediction. It introduces the Tensorized Quadrangle Attention (TQA) mechanism, which operates directly in Cartesian space, enabling accurate modeling of many-body interactions without relying on spherical harmonics, leading to significant computational savings.

We further provide LiTEN-FF, a foundation model for biomolecular force fields, pretrained and fine-tuned to support a wide range of tasks in both vacuum and solvated environments.(https://github.com/lingcon01/LiTEN-FF/tree/master)