Rigid Protein-Protein Docking via Equivariant Elliptic-Paraboloid Interface Prediction

Authors: Ziyang Yu, Wenbing Huang, Yang Liu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental evaluations show that Elli Dock achieves the fastest inference time among all compared methods and is strongly competitive with current state-of-the-art learning-based models such as Diff Dock-PP and Multimer particularly for antibody-antigen docking.
Researcher Affiliation Academia Ziyang Yu1 Wenbing Huang3,4, Yang Liu1,2, 1Department of Computer Science, Tsinghua University 2Institute for AI Industry Research (AIR), Tsinghua University 3Gaoling School of Artificial Intelligence, Renmin University of China 4Beijing Key Laboratory of Big Data Management and Analysis Methods, Beijing, China
Pseudocode No The paper describes its methodology using mathematical formulations and descriptive text, but it does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Our source code is available at https://github.com/yaledeus/Elli Dock.
Open Datasets Yes Datasets We utilize the following two datasets for our experiments: Docking benchmark version 5. We first leverage the Docking benchmark version 5 database (DB5.5) (Vreven et al., 2015), which is a gold standard dataset with 253 high-quality complex structures. We use the data split provided by Equi Dock, with partitioned sizes 203/25/25 for train/validation/test. The Structural Antibody Database. To further simulate real drug design scenarios and study the performance of our method on heterodimers, we select the Structural Antibody Database (SAb Dab) (Dunbar et al., 2014) for training. SAb Dab is a database of thousands of antibody-antigen complex structures that updates on a weekly basis.
Dataset Splits Yes We use the data split provided by Equi Dock, with partitioned sizes 203/25/25 for train/validation/test. Datasets are split based on sequence similarity measured by MMseqs2 (Steinegger & Söding, 2017), with partitioned sizes 1,781/300/0 for train/validation/test.
Hardware Specification Yes Our models are trained and tested on an Intel(R) Xeon(R) Gold 5218 CPU @ 2.30GHz and a NVIDIA Ge Force RTX 2080 Ti GPU.
Software Dependencies No The paper mentions the use of deep learning models and baselines, but it does not provide specific version numbers for software dependencies or libraries (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes Table 4: Hyperparameters of the training process. The bold marks represent the optimal parameters among their candidates based on validation results. Params DB5.5 SAb Dab learning rate 2e-4 2e-4, 3e-4, 5e-4 layers of EPIT L 2, 3 2, 3, 4, 5 embed size D 64 64 hidden size H 128 128 attention heads U 16 4, 8, 16, 20 neighbors 10 10 RBF size 20 20 radial cut ( A) 3.0 1.0, 2.0, 3.0 dropout rate 0.1 0.1