Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
SAGEPhos: Sage Bio-Coupled and Augmented Fusion for Phosphorylation Site Detection
Authors: Jingjie Zhang, Hanqun Cao, Zijun Gao, Xiaorui Wang, Chunbin Gu
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that SAGEPhos significantly outperforms baseline methods, notably achieving almost 10% and 12% improvements in prediction accuracy and AUC-ROC, respectively. We further demonstrate our algorithm s robustness and generalization through stable results across varied data partitions and significant improvements in zero-shot scenarios. These results underscore the effectiveness of constructing a larger and more precise protein space in advancing the state-of-the-art in phosphorylation site prediction. |
| Researcher Affiliation | Academia | Jingjie Zhang1, Hanqun Cao1, Zijun Gao1, Xiaorui Wang2, Chunbin Gu1 1Department of Computer Science and Engineering, The Chinese University of Hong Kong 2College of Pharmaceutical Sciences, Zhejiang University EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the model architecture and mechanisms using mathematical equations and diagrams (e.g., Figure 2, Figure 3) but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | We release the SAGEPhos models and code at https://github.com/ZhangJJ26/SAGEPhos. |
| Open Datasets | Yes | Our phosphorylation site prediction dataset was compiled from Phospho.ELM (Dinkel et al., 2010), Phospho Networks (Hu et al., 2014), and Phospho Site Plus (Hornbeck et al., 2012), with structural information incorporated from Alphafold DB (Jumper et al.). |
| Dataset Splits | Yes | To facilitate robust model training and evaluation, we partitioned the dataset into training, validation, and test sets using a ratio of 8:1:1. This approach ensures comprehensive assessment of our model s performance and generalization. Additionally, we modified the split methods to simulate a cold-start scenario (Zhu et al., 2021), introducing entirely new kinase or substrate sequences in the test set, denoted respectively as Kinase-cold-start and Substrate-cold-start. |
| Hardware Specification | Yes | Our experiments were implemented using Py Torch 2.2.2, leveraging an Intel(R) Xeon(R) Gold 6426Y CPU and 2 NVIDIA A40 GPUs for computational resources. |
| Software Dependencies | Yes | Our experiments were implemented using Py Torch 2.2.2, leveraging an Intel(R) Xeon(R) Gold 6426Y CPU and 2 NVIDIA A40 GPUs for computational resources. |
| Experiment Setup | Yes | We set the learning rate to 1e-5 and weight decay to 1e-4, with training conducted over 100 epochs. The predictor, a Multi-Layer Perceptron (MLP), consists of 3 layers with a dropout rate of 0.2. The learnable weights for both the gated module and residual module α1, α2, β1, and β2 range from 0.1 to 1.0. |