Injecting Multimodal Information into Rigid Protein Docking via Bi-level Optimization

Authors: Ruijia Wang, YiWu Sun, Yujie Luo, Shaochuan Li, Cheng Yang, Xingyi Cheng, Hui Li, Chuan Shi, Le Song

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on diverse datasets and evaluation protocols validate the effectiveness of Bi Dock. Compared to state-of-the-art baselines, Bi Dock achieves the maximum 234% relative improvement in challenging but practical antibody-antigen docking.
Researcher Affiliation Collaboration Ruijia Wang12 Yiwu Sun1 Yujie Luo1 Shaochuan Li1 Cheng Yang2 Xingyi Cheng1 Hui Li1 Chuan Shi2 Le Song1 1 Bio Map Research 2 Beijing University of Posts and Telecommunications {wangruijia, yangcheng, shichuan}@bupt.edu.cn, {yiwu, luoyujie, shaochuan, xingyi, lihui, songle}@biomap.com
Pseudocode No The paper contains mathematical formulations and descriptions but no explicit pseudocode or algorithm blocks.
Open Source Code No The paper mentions utilizing pretrained models for baselines (Equi Dock, Multimer) available on GitHub, but it does not provide a link or explicit statement about the open-source code for its proposed method, Bi Dock.
Open Datasets Yes We leverage Docking Benchmark 5.5 (DB5.5) [52], a gold standard dataset tailored for rigid docking. For a comprehensive comparison, we curate two datasets of antibodies (VH-VL) and antibody-antigen complexes (AB-AG) from Protein Data Bank (PDB) [3] and expect them to become new benchmarks.
Dataset Splits No The paper describes training and test set partitions ('the training set consists of 4,890 complexes', 'the test set comprises 68 antibody-antigen complexes') but does not explicitly state the use or size of a validation set split for hyperparameter tuning or early stopping.
Hardware Specification Yes The environment where we run all experiments is: Operating system: Linux version 5.13.0-30-generic CPU information: AMD EPYC 7742 64-Core Processor GPU information: NVIDIA A100-SXM4-80GB
Software Dependencies No The paper mentions using the 'Torch Opt' library and 'Multimer' checkpoint, but it does not specify version numbers for these or any other key software dependencies required for replication.
Experiment Setup Yes The crop size is set to 412, and the batch size is set to 1. The coefficients in Equation (13) are λ1 = 0.2, λ2 = 2.0, and λ3 = 10.0. For optimization, we employ the Adam optimizer with a learning rate of 10 4 and integrate learning rate warmup, gradually increasing the learning rate from 0 to 10 4 within the first 100 steps. The exponential moving average (EMA) strategy applies a decay rate of β = 0.999 and undergoes updates every 200 steps.