Protein-ligand binding representation learning from fine-grained interactions

Authors: Shikun Feng, Minghao Li, Yinjun Jia, Wei-Ying Ma, Yanyan Lan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments have demonstrated the superiority of our method across various binding tasks, including protein-ligand affinity prediction, virtual screening and protein-ligand docking.
Researcher Affiliation Academia Shikun Feng1 Minghao Li2 Yinjun Jia3 Weiying Ma1 Yanyan Lan1,4 1Institute for AI Industry Research, Tsinghua University 2Beijing Institute of Genomics, Chinese Academy of Sciences 3School of Life Sciences, Tsinghua University 4Beijing Academy of Artificial Intelligence (BAAI)
Pseudocode Yes Algorithm 1 Data Augmentation of ligand conformation
Open Source Code No The paper does not provide a direct link to a code repository or an explicit statement about releasing the source code for the methodology described.
Open Datasets Yes Bind Net is pre-trained on Bio Lip (Yang et al., 2012), where we solely use the entries for regular ligands. [...] The protein-ligand complexes and their associated binding strengths are obtained from the PDBBind dataset (Wang et al., 2005). [...] DUD-E (Mysinger et al., 2012) is a widely used benchmark for virtual screening [...]. [...] To validate our assumption, we conduct further experiments on the AD dataset (Chen et al., 2019)...
Dataset Splits Yes Finally, we report the testing results based on the model that yields the best validation performance on the validation set.
Hardware Specification Yes The model is trained for 30 epochs with a batch size of 32 batch size, which is completed on a machine equipped with 8-A100 GPUs.
Software Dependencies No The paper mentions software like 'RDKit', 'Uni-Mol encoders', 'Torch MD-NET', 'Frad', 'Coord', 'ESM2', and 'Pro FSA' but does not provide specific version numbers for any of them.
Experiment Setup Yes The model optimization is carried out using the Adam optimizer with a learning rate of 1e-4 and a weight decay of 1e-4. The mask ratio of MLR is set to 0.8, and ADP and MLR losses are treated equally. The model is trained for 30 epochs with a batch size of 32 batch size...