Antigen-Specific Antibody Design via Direct Energy-based Preference Optimization
Authors: Xiangxin Zhou, Dongyu Xue, Ruizhe Chen, Zaixiang Zheng, Liang Wang, Quanquan Gu
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on RAb D benchmark show that our approach effectively optimizes the energy of generated antibodies and achieves state-of-the-art performance in designing high-quality antibodies with low total energy and high binding affinity simultaneously, demonstrating the superiority of our approach. |
| Researcher Affiliation | Collaboration | Xiangxin Zhou1,2,4, Dongyu Xue4, Ruizhe Chen3,4, Zaixiang Zheng4 Liang Wang1,2 Quanquan Gu4, 1School of Artificial Intelligence, University of Chinese Academy of Sciences 2New Laboratory of Pattern Recognition (NLPR), State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of Automation, Chinese Academy of Sciences (CASIA) 3College of Computer Science and Electronic Engineering, Hunan University 4Byte Dance Research |
| Pseudocode | No | The paper describes its method in prose and mathematical equations, but does not include explicit 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | Answer: [No] Justification: The release of code requires approval. |
| Open Datasets | Yes | Dataset Curation To pre-train the diffusion model for antibody generation, we use the Structural Antibody Database [SAb Dab, 13] under IMGT [34] scheme as the dataset. |
| Dataset Splits | Yes | We then select the clusters that do not contain complexes in RAb D benchmark [1] and split the complexes into training and validation sets with a ratio of 9:1 (1786 and 193 complexes respectively). |
| Hardware Specification | Yes | We trained the model on one NVIDIA A100 80G GPU and it could converge within 30 hours and 200k steps. |
| Software Dependencies | No | The paper mentions software tools such as 'py Rosetta' [9], 'Rosetta', and 'MMseqs2' [44] but does not provide specific version numbers for these dependencies, which is necessary for reproducibility. |
| Experiment Setup | Yes | Pre-training Following Luo et al. [36], the diffusion model is first trained via the gradient descent method Adam [27] with init_learning_rate=1e-4, betas=(0.9,0.999), batch_size=16, and clip_gradient_norm=100. During the training phase, the weight of rotation loss, position loss, and sequence loss are each set to 1.0. We also schedule to decay the learning rate multiplied by a factor of 0.8 and a minimum learning rate of 5e 6. |