Unified Generative Modeling of 3D Molecules with Bayesian Flow Networks

Authors: Yuxuan Song, Jingjing Gong, Hao Zhou, Mingyue Zheng, Jingjing Liu, Wei-Ying Ma

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct thorough evaluations of Geo BFN on multiple benchmarks, including both unconditional and property-conditioned molecule generation tasks. Results demonstrate that Geo BFN consistently achieves state-of-the-art generation performance on molecule stability and other metrics.
Researcher Affiliation Academia 1 Institute of AI Industry Research (AIR), Tsinghua University 2 Shanghai Institute of Materia Medica, Chinese Academy of Sciences
Pseudocode Yes E DETAILED ALGORITHMS FOR TRAINING AND SAMPLING For a better understanding of the whole procedure in training and sampling, we involve the detailed algorithms and implements of functions in Algorithm 1, Algorithm 2 and Algorithm 3.
Open Source Code Yes 2The official implementation is at https://github.com/Algo Mole/Geo BFN
Open Datasets Yes The widely adapted QM9 (Ramakrishnan et al., 2014) and the GEOM-DRUG (Gebauer et al., 2019; 2021) with large molecules are used for the experiments.
Dataset Splits Yes The data configurations directly follow previous work(Anderson et al., 2019; Hoogeboom et al., 2022; Xu et al., 2023)2.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) were found for running its experiments.
Software Dependencies No The bayesian flow network is implemented with EGNNs Satorras et al. (2021b) by Py Torch (Paszke et al., 2017) package. We set the dimension of latent invariant features k to 1 for QM9 and 2 for DRUG, which extremely reduces the atomic feature dimension. For the training of vector field network vθ: on QM9, we train EGNNs with 9 layers and 256 hidden features with a batch size 64; and on DRUG, we train EGNNs with 4 layers and 256 hidden features, with batch size 64. The model uses Si LU activations. We train all the modules until convergence. For all the experiments, we choose the Adam optimizer (Kingma & Ba, 2014) with a constant learning rate of 10-4 as our default training configuration.
Experiment Setup Yes For the training of vector field network vθ: on QM9, we train EGNNs with 9 layers and 256 hidden features with a batch size 64; and on DRUG, we train EGNNs with 4 layers and 256 hidden features, with batch size 64. The model uses Si LU activations. We train all the modules until convergence. For all the experiments, we choose the Adam optimizer (Kingma & Ba, 2014) with a constant learning rate of 10 4 as our default training configuration.