Customized Subgraph Selection and Encoding for Drug-drug Interaction Prediction

Authors: Haotong Du, Quanming Yao, Juzheng Zhang, Yang Liu, Zhen Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness and superiority of the proposed method, with the discovered subgraphs and encoding functions highlighting the model s adaptability.
Researcher Affiliation Academia Haotong Du1 Quanming Yao2 Juzheng Zhang2 Yang Liu1 Zhen Wang1 1Northwestern Polytechnical University 2Tsinghua University
Pseudocode Yes Algorithm 1: The search algorithm of CSSE-DDI.
Open Source Code Yes Our code is available at https://github.com/LARS-research/CSSE-DDI.
Open Datasets Yes Experiments are conducted on two public benchmark DDI datasets: Drug Bank [42] and TWOSIDES [43]. Detailed descriptions of these datasets are presented in Appendix B.1.
Dataset Splits Yes Let Dtra and Dval denote the training and validation sets, respectively.
Hardware Specification Yes All the experiments are implemented in Python with the Py Torch framework [64] and run on a server machine with single NVIDIA RTX 3090 GPU with 24GB memory and 64GB of RAM.
Software Dependencies Yes All the experiments are implemented in Python with the Py Torch framework [64]
Experiment Setup Yes For CSSE-DDI, we set the epoch to 400 for training supernet and set the epoch to 400 for training sub-supernets. We set the the temperature parameter as 0.05. Repeat 5 times with different seeds, we can get 5 candidates. The searched candidates are finetuned individually with the hyper-parameters. In the stage of fine-tuning, we use the Reduce LROn Plateau scheduler to adjust the learning rate dynamically. Each candidate has 10 hyper steps. In each hyper step, a set of hyperparameter will be sampled from Table 8.