Geometry Informed Tokenization of Molecules for Language Model Generation
This addresses the challenge of tokenizing 3D molecular geometries for AI-driven drug discovery, representing an incremental advance by adapting existing tokenization concepts to a new domain.
The paper tackles the problem of generating 3D molecular geometries using language models by proposing Geo2Seq, a method that converts geometries into SE(3)-invariant discrete sequences, enabling various LMs to excel in molecular geometry generation, particularly in controlled tasks.
We consider molecule generation in 3D space using language models (LMs), which requires discrete tokenization of 3D molecular geometries. Although tokenization of molecular graphs exists, that for 3D geometries is largely unexplored. Here, we attempt to bridge this gap by proposing the Geo2Seq, which converts molecular geometries into $SE(3)$-invariant 1D discrete sequences. Geo2Seq consists of canonical labeling and invariant spherical representation steps, which together maintain geometric and atomic fidelity in a format conducive to LMs. Our experiments show that, when coupled with Geo2Seq, various LMs excel in molecular geometry generation, especially in controlled generation tasks.