v1.0.0: VGAE applied to GM12878 vs IMR90 chr21 Hi-C at 25kb

Full reproducible pipeline: .mcool + ChIP-seq bigwigs → latent
  embeddings → A/B compartment calls → cross-cell comparison.

  Key results (chr21, 25 kb, latent dim=32):
  - Test AUC=0.777, AP=0.759 (converged epoch 31/300)
  - GM12878 A/B silhouette (cosine) = 0.775
  - IMR90 zero-shot silhouette = 0.443
  - A-compartment bins stable across cell types (mean cosine Δ=0.042)
  - B-compartment bins shift substantially (mean cosine Δ=0.451)
  - 101 B→A and 70 A→B compartment switches GM12878→IMR90

env.yml

@@ -1,21 +1,27 @@
-name: chromatin_gnn_aman
+name: chromatin_gnn
+# Installation (pip-based, conda used only for Python):
+#   conda create -n chromatin_gnn python=3.10 -y
+#   conda activate chromatin_gnn
+#   pip install torch==2.1.2 --index-url https://download.pytorch.org/whl/cpu
+#   pip install -r requirements.txt
+#
+# For GPU support replace the torch line with:
+#   pip install torch==2.1.2 --index-url https://download.pytorch.org/whl/cu118
 channels:
-  - pytorch
-  - conda-forge
   - defaults
 dependencies:
   - python=3.10
-  - pytorch
-  - torchvision
-  - torchaudio
-  - cooler
-  - pybigwig
-  - pandas
-  - numpy
-  - scikit-learn
-  - matplotlib
-  - umap-learn
   - pip
   - pip:
-      - torch-geometric
-
+    - torch==2.1.2           # CPU build; see GPU note above
+    - torch-geometric==2.5.3
+    - cooler==0.9.3
+    - pyBigWig==0.3.25
+    - pandas==2.1.4
+    - "numpy>=1.24,<2.0"     # cooler 0.9.3 requires numpy<2.0
+    - scikit-learn==1.4.2
+    - matplotlib==3.8.4
+    - umap-learn==0.5.12
+    - scipy==1.12.0
+    - seaborn==0.13.2
+    - tqdm==4.66.2