Global Context-aware Representation Learning for Spatially Resolved Transcriptomics
Authors: Yunhak Oh, Junseok Lee, Yeongmin Kim, Sangwoo Seo, Namkyeong Lee, Chanyoung Park
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the superiority of Spotscape in various downstream tasks, including single-slice and multi-slice scenarios. |
| Researcher Affiliation | Academia | 1Graduate School of Data Science, KAIST, Daejeon, Republic of Korea 2Department of Industrial & Systems Engineering, KAIST, Daejeon, Republic of Korea 3Department of Mathematical Sciences, KAIST, Daejeon, Republic of Korea. Correspondence to: Chanyoung Park <EMAIL>. |
| Pseudocode | Yes | C. Pseudo Code In this section, we provide pseudocode of Spotscape in Algorithm 1. |
| Open Source Code | Yes | Our code is available at the following link: https: //github.com/yunhak0/Spotscape. |
| Open Datasets | Yes | We conduct a comprehensive evaluation of Spotscape across five datasets derived from different sequencing technologies. For single-slice experiments, we use the dorsolateral prefrontal cortex (DLPFC) dataset... Additionally, we assess the middle temporal gyrus (MTG) dataset... as well as the Mouse embryo dataset. Lastly, we utilize non-small cell lung cancer (NSCLC) data... In multi-slice experiments, we integrate the four slices from the same patient in the DLPFC dataset... while analyzing the differences between the control and AD groups in the MTG dataset... Lastly, we evaluate heterogeneous alignment using the Mouse embryo dataset... and the Breast Cancer dataset... Further details about data statistics can be found in Table 5 of Appendix A. |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits or specific percentages/counts for the datasets used in its experiments. It mentions using K-means clustering for spatial domain identification, which is an unsupervised task, and describes multi-slice integration/alignment, but no detailed splits for model training or evaluation phases. |
| Hardware Specification | Yes | All the experiments are conducted on Intel Xeon Gold 6326 CPU and NVIDIA Ge Force A6000 (48GB). |
| Software Dependencies | Yes | Spotscape is implemented in Python 3 (version 3.9.7) using Py Torch 2.1.1 (https: //pytorch.org/) with Pytorch Geometric (https://github.com/pyg-team/pytorch_geometric) packages. |
| Experiment Setup | Yes | To ensure a fair comparison, we conducted a hyperparameter search for both Spotscape and the baseline methods. The best-performing hyperparameters were selected by evaluating the NMI with the first seed. Specifically, for Spotscape, the hyperparameter search was conducted only for the learning rate, with the search space consisting of {0.00001, 0.00005, 0.0001, 0.0005, 0.001}. The remaining hyperparameters were fixed, and the ones used to report the experimental results are listed in Table 7. Table 7. Hyperparameter settings of Spotscape. (lists specific values for λRecon, λSC, λPCL, λSS, GCN encoder dimensions, τ, Top-k, Training epochs, Warm-up epochs, Learning rate, Feature masking rate, Edge masking rate) |