sphot 0.1.0__tar.gz

sphot-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Hoeyoung Kim and Donghee Kim
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

sphot-0.1.0/PKG-INFO

@@ -0,0 +1,177 @@
+Metadata-Version: 2.4
+Name: sphot
+Version: 0.1.0
+Summary: Phenotype-associated spatial biomarker discovery in spatial transcriptomics with spHOT
+Author: Hoeyoung Kim, Donghee Kim
+License: MIT License
+        
+        Copyright (c) 2026 Hoeyoung Kim and Donghee Kim
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+        
+Project-URL: Homepage, https://github.com/DHKim327/spHOT
+Project-URL: Repository, https://github.com/DHKim327/spHOT
+Keywords: spatial transcriptomics,multiple instance learning,spatial biomarker,bioinformatics
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Operating System :: POSIX :: Linux
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.23
+Requires-Dist: pandas>=1.5
+Requires-Dist: scipy>=1.10
+Requires-Dist: scikit-learn>=1.2
+Requires-Dist: scanpy>=1.9
+Requires-Dist: anndata>=0.9
+Requires-Dist: torch>=2.0
+Requires-Dist: novae>=1.0.0
+Requires-Dist: networkx>=3.0
+Requires-Dist: matplotlib>=3.6
+Requires-Dist: seaborn>=0.12
+Requires-Dist: termcolor>=2.0
+Requires-Dist: pynvml>=11.0
+Dynamic: license-file
+
+# spHOT
+### Phenotype-associated spatial biomarker discovery in spatial transcriptomics with spHOT
+[![License: MIT](https://img.shields.io/badge/license-MIT-pink.svg)](https://github.com/DHKim327/spHOT/blob/main/LICENSE)
+[![Python](https://img.shields.io/badge/python-3.10-pink.svg)](https://www.python.org/)
+
+
+## 🧬 Description
+**🔥spHOT🔥** is a framework for localizing phenotype-associated spatial biomarkers from multi-sample, multi-patient spatial transcriptomics datasets with sample-level case/control labels. spHOT integrates **spatial foundation model embeddings**, a **hierarchical domain tree**, and a dual-branch **teacher–student multiple instance learning architecture** to convert sample-level phenotype labels into spatially coherent cell-level biomarker scores.
+
+<img src="https://raw.githubusercontent.com/DHKim327/spHOT/main/docs/spHOT_Figure1_V3.png" width="1000px" align="center" />
+
+
+## ⚙️ Installation
+
+**Option 1 — PyPI**
+```bash
+pip install sphot
+```
+
+**Option 2 — GitHub**
+```bash
+pip install git+https://github.com/DHKim327/spHOT.git
+```
+
+**Option 3 — conda yml**
+```bash
+git clone https://github.com/DHKim327/spHOT.git
+cd spHOT
+conda env create -f environment/env_spHOT.yml
+conda activate env_spHOT
+pip install -e .
+```
+
+## 📥 Inputs
+
+**AnnData directory :**
+<br>Path to `.h5ad` files, saved per sample :
+```
+adatas/
+├── sample1/adata.h5ad
+└── sample2/adata.h5ad
+```
+- `.obs` key should contain sample phenotype label
+
+**Split information :**
+
+The `splits.csv` file defines train/valid/test splits for each sample across multiple folds.
+
+| SID         | 0     | 1     | 2     | 3     | 4     |
+|--------------|-------|-------|-------|-------|-------|
+| sample_1     | train | train | train | test  | valid |
+| sample_2     | train | test  | valid | train | train |
+| sample_3     | valid | train | train | train | test  |
+
+- **SID :** Sample identifier (e.g., slide or tissue ID)  
+- **0–4 :** Fold indices  
+- Each cell indicates the dataset role of that sample for a specific fold.
+
+
+## 📤 Outputs
+
+After an spHOT run, the output directory structure will be organized as below :
+```
+results/
+└── {task_name}/                          # Name of run
+    ├── de/                               # Domain embedding module result
+    │   └── adata.h5ad                    # Merged AnnData with Novae embeddings & initial domain info           
+    ├── dt/                               # Domain tree module result
+    |   ├── centroid_HC_results.pkl       # Hierarchical domain tree
+    |   └── adata.h5ad                    # Merged AnnData with metadomains association info
+    └── mil/                              # MIL module result
+        ├── D{k}/                         # k-level MIL result
+        │   ├── model_encoder_exp{exp}.pt     # Model and outputs for each fold
+        │   ├── model_teacher_exp{exp}.pt             
+        │   ├── model_student_exp{exp}.pt            
+        │   ├── resource_{exp}.csv            # Resource performance log
+        │   └── CELL_SCORE_test_{exp}.h5ad    # spHOT cell scores (test) for each fold
+        ├── all_test_results.csv          # Sample classification performance of test samples
+        ├── all_test_results.csv          # Sample classification performance of validation samples
+        └── k_selection_results.csv       # Spatial scores and k-selection result of spHOT run
+```
+
+## 📘 Tutorials
+
+We provide the code and resources required to reproduce all main figures in the manuscript within the `./tutorial` directory. Also, see the `./tutorial` directory for step-by-step spHOT usage.
+
+### Figure-to-notebook map
+
+| Figure | Dataset | Step | Notebook |
+|---|---|---|---|
+| **Fig. 2+@** | Simulation | Source data preparation | [`0.preproc_CosMx_Pouch_IBD.ipynb`](./tutorial/Fig2_Simulation/0.preproc_CosMx_Pouch_IBD.ipynb) |
+| | | Simulation generation (V1) | [`1.simul_scCube_V1.ipynb`](./tutorial/Fig2_Simulation/1.simul_scCube_V1.ipynb) |
+| | | Simulation generation (V2) | [`1.simul_scCube_V2.ipynb`](./tutorial/Fig2_Simulation/1.simul_scCube_V2.ipynb) |
+| | | spHOT run (V1) | [`2.spHOT_simul_V1.ipynb`](./tutorial/Fig2_Simulation/2.spHOT_simul_V1.ipynb) |
+| | | spHOT run (V2) | [`2.spHOT_simul_V2.ipynb`](./tutorial/Fig2_Simulation/2.spHOT_simul_V2.ipynb) |
+| | | Evaluation (V1) | [`3.eval_simul_V1.ipynb`](./tutorial/Fig2_Simulation/3.eval_simul_V1.ipynb) |
+| | | Evaluation (V2) | [`3.eval_simul_V2.ipynb`](./tutorial/Fig2_Simulation/3.eval_simul_V2.ipynb) |
+| **Fig. 3–4** | Xenium IPF (GSE250346) | Preprocessing | [`0.preproc_Xenium_IPF.R`](./tutorial/Fig3_4_Xenium_IPF/0.preproc_Xenium_IPF.R) · [`0.preproc_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/0.preproc_Xenium_IPF.ipynb) |
+| | | spHOT run | [`1.spHOT_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/1.spHOT_Xenium_IPF.ipynb) |
+| | | Evaluation | [`2.eval_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/2.eval_Xenium_IPF.ipynb) |
+| | | Downstream analysis | [`3.downstream_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/3.downstream_Xenium_IPF.ipynb) |
+| **Fig. 5** | CosMx DKD | Preprocessing | [`0.preproc_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/0.preproc_CosMx_DKD.ipynb) |
+| | | spHOT run | [`1.spHOT_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/1.spHOT_CosMx_DKD.ipynb) |
+| | | Evaluation | [`2.eval_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/2.eval_CosMx_DKD.ipynb) |
+| | | Downstream analysis | [`3.downstream_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/3.downstream_CosMx_DKD.ipynb) |
+| **Fig. 6** | Xenium RPGN (GSE294965) | Preprocessing | [`0.preproc_Xenium_RPGN.ipynb`](./tutorial/Fig6_Xenium_RPGN/0.preproc_Xenium_RPGN.ipynb) |
+| | | spHOT run | [`1.spHOT_Xenium_RPGN.ipynb`](./tutorial/Fig6_Xenium_RPGN/1.spHOT_Xenium_RPGN.ipynb) |
+| | | Downstream analysis (1) | [`2.downstream_Xenium_RPGN_V1.ipynb`](./tutorial/Fig6_Xenium_RPGN/2.downstream_Xenium_RPGN_V1.ipynb) |
+| | | Downstream analysis (2) | [`2.downstream_Xenium_RPGN_V2.ipynb`](./tutorial/Fig6_Xenium_RPGN/2.downstream_Xenium_RPGN_V2.ipynb) |
+| **Fig. 7** | Xenium COPD (GSE313006) | Preprocessing & cell typing | [`0.preproc_Xenium_COPD.ipynb`](./tutorial/Fig7_Xenium_COPD/0.preproc_Xenium_COPD.ipynb) |
+| | | Downstream analysis (1) | [`1.downstream_Xenium_COPD_V1.ipynb`](./tutorial/Fig7_Xenium_COPD/1.downstream_Xenium_COPD_V1.ipynb) |
+| | | Downstream analysis (2) | [`1.downstream_Xenium_COPD_V2.ipynb`](./tutorial/Fig7_Xenium_COPD/1.downstream_Xenium_COPD_V2.ipynb) |
+| **Fig. 2, 3, 5+@** | All datasets | Cross-dataset benchmarking & statistics | [`benchmark_stats_all_datasets.ipynb`](./tutorial/benchmark_stats_all_datasets.ipynb) |
+
+> **Note :** Notebooks are numbered by execution order within each figure directory (`0.` → `3.`). Preprocessing notebooks start from public GEO accessions; the simulation dataset is downloaded from Zenodo as the version of record (see [Data availability](https://doi.org/10.5281/zenodo.21156231)).
+
+## 😊 Acknowledgements
+spHOT is built upon codes from [scMILD: Single-cell multiple instance learning for sample classification and associated subpopulation discovery](https://github.com/Khreat0205/scMILD). We thank the authors for publicly releasing their codes.
+
+
+## 📚 References
+Jeong, K., Choi, J. & Kim, K. scMILD: Single-cell multiple instance learning for sample classification and associated subpopulation discovery. iScience 29(2026).
+
+
+<br><br>

sphot-0.1.0/README.md

@@ -0,0 +1,125 @@
+# spHOT
+### Phenotype-associated spatial biomarker discovery in spatial transcriptomics with spHOT
+[![License: MIT](https://img.shields.io/badge/license-MIT-pink.svg)](https://github.com/DHKim327/spHOT/blob/main/LICENSE)
+[![Python](https://img.shields.io/badge/python-3.10-pink.svg)](https://www.python.org/)
+
+
+## 🧬 Description
+**🔥spHOT🔥** is a framework for localizing phenotype-associated spatial biomarkers from multi-sample, multi-patient spatial transcriptomics datasets with sample-level case/control labels. spHOT integrates **spatial foundation model embeddings**, a **hierarchical domain tree**, and a dual-branch **teacher–student multiple instance learning architecture** to convert sample-level phenotype labels into spatially coherent cell-level biomarker scores.
+
+<img src="https://raw.githubusercontent.com/DHKim327/spHOT/main/docs/spHOT_Figure1_V3.png" width="1000px" align="center" />
+
+
+## ⚙️ Installation
+
+**Option 1 — PyPI**
+```bash
+pip install sphot
+```
+
+**Option 2 — GitHub**
+```bash
+pip install git+https://github.com/DHKim327/spHOT.git
+```
+
+**Option 3 — conda yml**
+```bash
+git clone https://github.com/DHKim327/spHOT.git
+cd spHOT
+conda env create -f environment/env_spHOT.yml
+conda activate env_spHOT
+pip install -e .
+```
+
+## 📥 Inputs
+
+**AnnData directory :**
+<br>Path to `.h5ad` files, saved per sample :
+```
+adatas/
+├── sample1/adata.h5ad
+└── sample2/adata.h5ad
+```
+- `.obs` key should contain sample phenotype label
+
+**Split information :**
+
+The `splits.csv` file defines train/valid/test splits for each sample across multiple folds.
+
+| SID         | 0     | 1     | 2     | 3     | 4     |
+|--------------|-------|-------|-------|-------|-------|
+| sample_1     | train | train | train | test  | valid |
+| sample_2     | train | test  | valid | train | train |
+| sample_3     | valid | train | train | train | test  |
+
+- **SID :** Sample identifier (e.g., slide or tissue ID)  
+- **0–4 :** Fold indices  
+- Each cell indicates the dataset role of that sample for a specific fold.
+
+
+## 📤 Outputs
+
+After an spHOT run, the output directory structure will be organized as below :
+```
+results/
+└── {task_name}/                          # Name of run
+    ├── de/                               # Domain embedding module result
+    │   └── adata.h5ad                    # Merged AnnData with Novae embeddings & initial domain info           
+    ├── dt/                               # Domain tree module result
+    |   ├── centroid_HC_results.pkl       # Hierarchical domain tree
+    |   └── adata.h5ad                    # Merged AnnData with metadomains association info
+    └── mil/                              # MIL module result
+        ├── D{k}/                         # k-level MIL result
+        │   ├── model_encoder_exp{exp}.pt     # Model and outputs for each fold
+        │   ├── model_teacher_exp{exp}.pt             
+        │   ├── model_student_exp{exp}.pt            
+        │   ├── resource_{exp}.csv            # Resource performance log
+        │   └── CELL_SCORE_test_{exp}.h5ad    # spHOT cell scores (test) for each fold
+        ├── all_test_results.csv          # Sample classification performance of test samples
+        ├── all_test_results.csv          # Sample classification performance of validation samples
+        └── k_selection_results.csv       # Spatial scores and k-selection result of spHOT run
+```
+
+## 📘 Tutorials
+
+We provide the code and resources required to reproduce all main figures in the manuscript within the `./tutorial` directory. Also, see the `./tutorial` directory for step-by-step spHOT usage.
+
+### Figure-to-notebook map
+
+| Figure | Dataset | Step | Notebook |
+|---|---|---|---|
+| **Fig. 2+@** | Simulation | Source data preparation | [`0.preproc_CosMx_Pouch_IBD.ipynb`](./tutorial/Fig2_Simulation/0.preproc_CosMx_Pouch_IBD.ipynb) |
+| | | Simulation generation (V1) | [`1.simul_scCube_V1.ipynb`](./tutorial/Fig2_Simulation/1.simul_scCube_V1.ipynb) |
+| | | Simulation generation (V2) | [`1.simul_scCube_V2.ipynb`](./tutorial/Fig2_Simulation/1.simul_scCube_V2.ipynb) |
+| | | spHOT run (V1) | [`2.spHOT_simul_V1.ipynb`](./tutorial/Fig2_Simulation/2.spHOT_simul_V1.ipynb) |
+| | | spHOT run (V2) | [`2.spHOT_simul_V2.ipynb`](./tutorial/Fig2_Simulation/2.spHOT_simul_V2.ipynb) |
+| | | Evaluation (V1) | [`3.eval_simul_V1.ipynb`](./tutorial/Fig2_Simulation/3.eval_simul_V1.ipynb) |
+| | | Evaluation (V2) | [`3.eval_simul_V2.ipynb`](./tutorial/Fig2_Simulation/3.eval_simul_V2.ipynb) |
+| **Fig. 3–4** | Xenium IPF (GSE250346) | Preprocessing | [`0.preproc_Xenium_IPF.R`](./tutorial/Fig3_4_Xenium_IPF/0.preproc_Xenium_IPF.R) · [`0.preproc_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/0.preproc_Xenium_IPF.ipynb) |
+| | | spHOT run | [`1.spHOT_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/1.spHOT_Xenium_IPF.ipynb) |
+| | | Evaluation | [`2.eval_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/2.eval_Xenium_IPF.ipynb) |
+| | | Downstream analysis | [`3.downstream_Xenium_IPF.ipynb`](./tutorial/Fig3_4_Xenium_IPF/3.downstream_Xenium_IPF.ipynb) |
+| **Fig. 5** | CosMx DKD | Preprocessing | [`0.preproc_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/0.preproc_CosMx_DKD.ipynb) |
+| | | spHOT run | [`1.spHOT_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/1.spHOT_CosMx_DKD.ipynb) |
+| | | Evaluation | [`2.eval_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/2.eval_CosMx_DKD.ipynb) |
+| | | Downstream analysis | [`3.downstream_CosMx_DKD.ipynb`](./tutorial/Fig5_CosMx_DKD/3.downstream_CosMx_DKD.ipynb) |
+| **Fig. 6** | Xenium RPGN (GSE294965) | Preprocessing | [`0.preproc_Xenium_RPGN.ipynb`](./tutorial/Fig6_Xenium_RPGN/0.preproc_Xenium_RPGN.ipynb) |
+| | | spHOT run | [`1.spHOT_Xenium_RPGN.ipynb`](./tutorial/Fig6_Xenium_RPGN/1.spHOT_Xenium_RPGN.ipynb) |
+| | | Downstream analysis (1) | [`2.downstream_Xenium_RPGN_V1.ipynb`](./tutorial/Fig6_Xenium_RPGN/2.downstream_Xenium_RPGN_V1.ipynb) |
+| | | Downstream analysis (2) | [`2.downstream_Xenium_RPGN_V2.ipynb`](./tutorial/Fig6_Xenium_RPGN/2.downstream_Xenium_RPGN_V2.ipynb) |
+| **Fig. 7** | Xenium COPD (GSE313006) | Preprocessing & cell typing | [`0.preproc_Xenium_COPD.ipynb`](./tutorial/Fig7_Xenium_COPD/0.preproc_Xenium_COPD.ipynb) |
+| | | Downstream analysis (1) | [`1.downstream_Xenium_COPD_V1.ipynb`](./tutorial/Fig7_Xenium_COPD/1.downstream_Xenium_COPD_V1.ipynb) |
+| | | Downstream analysis (2) | [`1.downstream_Xenium_COPD_V2.ipynb`](./tutorial/Fig7_Xenium_COPD/1.downstream_Xenium_COPD_V2.ipynb) |
+| **Fig. 2, 3, 5+@** | All datasets | Cross-dataset benchmarking & statistics | [`benchmark_stats_all_datasets.ipynb`](./tutorial/benchmark_stats_all_datasets.ipynb) |
+
+> **Note :** Notebooks are numbered by execution order within each figure directory (`0.` → `3.`). Preprocessing notebooks start from public GEO accessions; the simulation dataset is downloaded from Zenodo as the version of record (see [Data availability](https://doi.org/10.5281/zenodo.21156231)).
+
+## 😊 Acknowledgements
+spHOT is built upon codes from [scMILD: Single-cell multiple instance learning for sample classification and associated subpopulation discovery](https://github.com/Khreat0205/scMILD). We thank the authors for publicly releasing their codes.
+
+
+## 📚 References
+Jeong, K., Choi, J. & Kim, K. scMILD: Single-cell multiple instance learning for sample classification and associated subpopulation discovery. iScience 29(2026).
+
+
+<br><br>

sphot-0.1.0/pyproject.toml

@@ -0,0 +1,49 @@
+[build-system]
+requires = ["setuptools>=61", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "sphot"
+version = "0.1.0"
+description = "Phenotype-associated spatial biomarker discovery in spatial transcriptomics with spHOT"
+readme = "README.md"
+requires-python = ">=3.10"
+license = { file = "LICENSE" }
+authors = [
+    { name = "Hoeyoung Kim" },
+    { name = "Donghee Kim" },
+]
+keywords = [
+    "spatial transcriptomics",
+    "multiple instance learning",
+    "spatial biomarker",
+    "bioinformatics",
+]
+classifiers = [
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Operating System :: POSIX :: Linux",
+]
+dependencies = [
+    "numpy>=1.23",
+    "pandas>=1.5",
+    "scipy>=1.10",
+    "scikit-learn>=1.2",
+    "scanpy>=1.9",
+    "anndata>=0.9",
+    "torch>=2.0",
+    "novae>=1.0.0",
+    "networkx>=3.0",
+    "matplotlib>=3.6",
+    "seaborn>=0.12",
+    "termcolor>=2.0",
+    "pynvml>=11.0",
+]
+
+[project.urls]
+Homepage = "https://github.com/DHKim327/spHOT"
+Repository = "https://github.com/DHKim327/spHOT"
+
+[tool.setuptools.packages.find]
+include = ["sphot*"]

sphot-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

sphot-0.1.0/sphot/de/__init__.py

@@ -0,0 +1 @@
+from .base import DomainEmbedder

sphot-0.1.0/sphot/de/base.py

@@ -0,0 +1,88 @@
+import os
+from pathlib import Path
+from ..utils import load_adatas, set_seed
+
+class DomainEmbedder:
+    def __init__(self, task_name,cfgs):
+        self.task_name = task_name
+        self.cfgs = cfgs
+        self.model_name = cfgs.get('model')  # e.g., 'Novae'
+        self.adata_dir = cfgs.get('adata_dir')
+        
+        self.save_dir = Path(cfgs.get('save_dir'))/ self.task_name / 'de'
+        self.save_dir.mkdir(parents=True, exist_ok=True)
+        
+        # Set GPU device (applies to all embedders)
+        params = cfgs.get('params', {})
+        resources = params.get('resources', {})
+        gpu_id = resources.get('gpu_id', None)
+        if gpu_id is not None:
+            self._set_gpu_device(gpu_id)
+        
+        # Set random seed for reproducibility (applies to all embedders)
+        self.random_seed = params.get('random_seed', 42)
+        set_seed(self.random_seed, deterministic=True)
+        print(f"[DomainEmbedder] Random seed set to {self.random_seed} for {self.model_name}")
+    
+    def _set_gpu_device(self, gpu_id):
+        """
+        Set GPU device for all frameworks (PyTorch, TensorFlow, JAX, etc.)
+        
+        Args:
+            gpu_id (int or str): GPU device ID (e.g., 0, 1, 2, or "0,1" for multiple GPUs)
+        """
+        gpu_id_str = str(gpu_id)
+        
+        # 1. Set CUDA_VISIBLE_DEVICES (works for all frameworks)
+        os.environ['CUDA_VISIBLE_DEVICES'] = gpu_id_str
+        print(f"[DomainEmbedder] CUDA_VISIBLE_DEVICES set to: {gpu_id_str}")
+        
+
+    def model_selection(self):
+        if self.model_name == 'Novae':
+            from .embedder_novae import NOVAEmbedder
+            return NOVAEmbedder
+        else:
+            raise KeyError("Invalid model name for DomainEmbedding: expected 'Novae'")
+
+    def run(self, adata_func=load_adatas):
+        embedder_cls = self.model_selection()
+        embedder = embedder_cls(self.cfgs.get('params'),self.save_dir)
+
+        adatas = adata_func(self.adata_dir)
+        # 각 embedder는 run(adatas) 수행 후 내부 상태에 결과가 있고,
+        # get_embed()로 {'obsm_key','obs_key','adata_integrated'} dict를 반환한다고 가정
+        embedder._run(adatas)
+        result = embedder._get_result()
+
+        adata = self._postprocessing(result)
+        adata.write_h5ad(self.save_dir / f'adata.h5ad',compression='gzip')
+
+
+    def _postprocessing(self, result):
+        """
+        result:
+          - 'adata_integrated': AnnData (통합)
+          - 'obsm_key': latent embedding key (ex: 'novae_latent')
+          - (optional) 'obs_key': cluster key (ex: 'novae_leaves')
+        """
+        adata = result['adata_integrated']
+        obsm_key = result['obsm_key']
+        adata = self._obsm_process(adata, obsm_key)
+
+        if 'obs_key' in result:
+            obs_key = result['obs_key']
+            adata = self._obs_process(adata, obs_key)
+
+        return adata
+
+    def _obsm_process(self, adata, obsm_key):
+        adata.obsm[f'X_{self.model_name}'] = adata.obsm[obsm_key]
+        return adata
+
+    def _obs_process(self, adata, obs_key):
+        # 클러스터 라벨을 0..C-1 정수로 맵핑해서 'Cluster_{model_name}'에 저장
+        cats = adata.obs[obs_key].astype('category').cat.categories.tolist()
+        mapping = {c: i for i, c in enumerate(sorted(cats))}
+        adata.obs[f'Cluster_{self.model_name}'] = adata.obs[obs_key].map(mapping)
+        return adata

sphot-0.1.0/sphot/de/embedder_novae/__init__.py

@@ -0,0 +1 @@
+from .embedder import NOVAEmbedder

sphot-0.1.0/sphot/de/embedder_novae/embedder.py

@@ -0,0 +1,184 @@
+import os
+from pathlib import Path
+import novae
+import  scanpy as sc
+
+class NOVAEmbedder():
+    """
+    NOVAE-based Domain Embedding (Official Hugging Face Method)
+    
+    Official usage:
+    https://huggingface.co/MICS-Lab/novae-human-0
+    
+    Supports:
+    - Spatial Transcriptomics (pretrained & finetune modes)
+    - Spatial Proteomics (pretrained mode only)
+    
+    Modes:
+    - 'pretrained': Zero-shot inference using pretrained model
+    - 'finetune': Fine-tune on user data then compute representations
+    
+    Important: NOVAE creates spatial neighbor graphs (required for MIL k optimization)
+    
+    Input: configuration files, adatas:list at runtime
+    Output: {
+        'obsm_key': 'novae_latent',
+        'obs_key': 'novae_leaves',
+        'adata_integrated': concatenated AnnData
+    }
+    """
+    def __init__(self, params,save_dir):
+        self.params = params
+        self.save_dir = save_dir
+        self._setup()
+
+    def _setup(self):
+        """Initialize parameters and download model if needed"""
+        self.data_type = self.params.get('data_type', 'transcriptomics')
+        self.mode = self.params.get('mode', 'pretrained')
+        
+        # Validate mode
+        if self.mode not in ['pretrained', 'finetune']:
+            raise ValueError(
+                f"\n{'='*80}\n"
+                f"Invalid mode: '{self.mode}'\n"
+                f"{'='*80}\n"
+                f"NOVAE supports:\n"
+                f"  - 'pretrained' (zero-shot inference)\n"
+                f"  - 'finetune' (fine-tune then compute representations)\n"
+                f"\nPlease check your configuration file.\n"
+                f"{'='*80}\n"
+            )
+        
+        # NOVAE supports transcriptomics (and proteomics for pretrained mode only)
+        if self.data_type not in ['transcriptomics', 'proteomics']:
+            raise ValueError(
+                f"\n{'='*80}\n"
+                f"Invalid data_type: '{self.data_type}'\n"
+                f"{'='*80}\n"
+                f"NOVAE supports:\n"
+                f"  - 'transcriptomics' (pretrained & finetune modes)\n"
+                f"  - 'proteomics' (pretrained mode only)\n"
+                f"\nPlease check your configuration file.\n"
+                f"{'='*80}\n"
+            )
+        
+        # Resources
+        self.resources = self.params.get('resources', {})
+        
+        # Setup model directory
+        model_dir = self.params.get('model_dir')
+
+        if model_dir is None:
+            # Auto-download from Hugging Face Hub and save locally
+            model_dir = self.save_dir / 'novae_pretrained'
+            model_dir.mkdir(parents=True, exist_ok=True)
+            
+            print(f"[NOVAE] Downloading from Hugging Face Hub (MICS-Lab/novae-human-0)...")
+            model = novae.Novae.from_pretrained("MICS-Lab/novae-human-0")
+            model.save_pretrained(model_dir.as_posix())
+            print(f"[NOVAE] Model saved to {model_dir}")
+            
+            self.model_dir = model_dir
+        else:
+            # Use user-specified directory
+            self.model_dir = Path(model_dir)
+            print(f"[NOVAE] Using model from {self.model_dir}")
+        
+        print(f"[NOVAE] Data type: {self.data_type}")
+        print(f"[NOVAE] Mode: {self.mode}")
+
+    def _inference(self):
+        """Zero-shot inference using pretrained model"""
+        accelerator = self.resources.get('accelerator', 'cpu')
+        num_workers = self.resources.get('num_workers', 4)
+        
+        print(f"[NOVAE] Running zero-shot inference...")
+        print(f"[NOVAE]   - accelerator: {accelerator}")
+        print(f"[NOVAE]   - num_workers: {num_workers}")
+
+        model = novae.Novae.from_pretrained(self.model_dir.as_posix())
+        model.compute_representations(
+            self.adatas,
+            zero_shot=True,
+            accelerator=accelerator,
+            num_workers=num_workers,
+        )
+        
+        print(f"[NOVAE] Inference completed")
+        
+
+    def _finetune(self):
+        """Fine-tune pretrained model and compute representations"""
+        accelerator = self.resources.get('accelerator', 'cpu')
+        num_workers = self.resources.get('num_workers', 4)
+
+        fit_param = self.params.get('fit', {})
+        
+        print(f"[NOVAE] Fine-tuning model...")
+        print(f"[NOVAE]   - accelerator: {accelerator}")
+        print(f"[NOVAE]   - num_workers: {num_workers}")
+        print(f"[NOVAE]   - fit parameters: {fit_param}")
+
+        model = novae.Novae.from_pretrained(self.model_dir.as_posix())
+
+        # fine-tune
+        print(f"[NOVAE] Starting fine-tuning...")
+        model.fine_tune(
+            self.adatas,
+            accelerator=accelerator,
+            num_workers=num_workers,
+            **fit_param,          # 여기에는 reference, max_epochs, patience, min_delta 등
+        )
+        print(f"[NOVAE] Fine-tuning completed")
+
+        # representation 계산
+        print(f"[NOVAE] Computing representations...")
+        model.compute_representations(
+            self.adatas,
+            accelerator=accelerator,
+            num_workers=num_workers,
+        )
+        print(f"[NOVAE] Representations computed")
+
+        if self.params.get('save', True):
+            save_path = self.save_dir / 'novae_finetuned'
+            model.save_pretrained(save_path.as_posix())
+            print(f"[NOVAE] Fine-tuned model saved to {save_path}")
+        
+
+    def _construct_graph_per_adata(self):
+        """Build spatial neighbor graph for each AnnData (required for MIL k optimization)"""
+        spatial_params = self.params.get('spatial_neighbors', {})
+        radius = spatial_params.get('radius', 150)  # Default radius for NOVAE
+        
+        print(f"[NOVAE] Building spatial neighbor graphs...")
+        print(f"[NOVAE]   - radius: {radius}")
+        novae.spatial_neighbors(self.adatas, radius=radius)
+        print(f"[NOVAE] Spatial graphs completed for {len(self.adatas)} samples")
+
+    def _get_result(self):
+        return {
+            'obsm_key': 'novae_latent',
+            'obs_key': 'novae_leaves',
+            'adata_integrated': sc.concat(self.adatas)
+        }
+
+    def _run(self, adatas):
+        """Main execution function"""
+        self.adatas = adatas
+        
+        print(f"[NOVAE] Starting domain embedding for {len(adatas)} samples")
+        
+        # 1. Build spatial graphs (required for MIL k optimization)
+        self._construct_graph_per_adata()
+        
+        # 2. Run model (pretrained or finetune)
+        if self.mode == 'finetune':
+            self._finetune()
+        elif self.mode == 'pretrained':
+            self._inference()
+        else:
+            raise ValueError(f"Invalid mode in NOVAEmbedder: expected 'finetune' or 'pretrained', got '{self.mode}'")
+        
+        print(f"[NOVAE] Domain embedding completed")

sphot-0.1.0/sphot/dt/__init__.py

@@ -0,0 +1 @@
+from .base import MetaDomain