Designing molecules with desirable physiochemical properties and functionalities is a long-standing challenge in chemistry, material science, and drug discovery. Recently, machine learning-based generative models have emerged as promising approaches for de novo molecule design. However, further refinement of methodology is highly desired as most existing methods lack unified modeling of 2D topology and 3D geometry information and fail to efficiently learn the structure-property relationship for molecule generation. Here we present MolCode, a roto-translation equivariant generative framework for Molecular graph-structure Co-design. In MolCode, 3D geometric information is leveraged for molecular 2D graph generation, which in turn helps guide the prediction of molecular 3D structure. Extensive experimental results show that MolCode outperforms previous methods on a series of tasks including de novo molecule design, targeted molecule discovery, and structure-based drug design. Particularly, MolCode not only consistently generates valid (99.95% Validity) and diverse (98.75% Uniqueness) molecular graphs/structures with desirable properties, but also generate drug-like molecules with high affinity to target proteins (61.8% high affinity ratio), which demonstrate MolCode’s potential applications in material design and drug discovery. Our investigation reveals that the 2D topology and 3D geometry contain intrinsically complementary information in molecule generation, and provides new insights into machine learning-based molecule representation and generation.

Download QM9 data from https://deepchemdata.s3-us-west-1.amazonaws.com/datasets/gdb9.tar.gz, unzip the file and put the file gdb9.sdf under the folder qm9/

The CrossDocked2020 datasets is public available at https://drive.google.com/drive/folders/1CzwxmTpjbrt83z_wBzcQncq84OVDPurM

conda env create -f molcode_env.yaml
conda activate molcode_env

git clone https://github.com/zaixizhang/MolCode.git
cd MolCode
sh start.sh