An AI-Guided Liver Metastasis Atlas Links Tumor Invasion Programs to Invasive Trophoblast Signatures

before running the contrastive learning script, please use preprocess.py to create the data.

python preprocess.py --dataset {dataset name} --dataset_key {batch name} --obs_fields {obs fields to be copied} --layer_fields {layer fields to be copied} --npy_dataset {npz output}

then use train_simclr.py

python train_simclr.py --train_data_name {data file} --embedding_save {embedding-output} --dataset_symbol {name of dataset field} --label_symbols {name of label field} --data_symbol {name of data field}

envrionment

The envrioment can be found in env.yml. The following python packages were used for training and prediction:

python=3.11.11

plotly==6.0.1 numpy==2.2.5 pandas==2.2.3 pillow==11.2.1 scanpy==1.11.1 scikit-learn==1.5.2 scikit-misc==0.5.1 scipy==1.15.2 scvi==0.6.8 scvi-tools==1.3.0 seaborn==0.13.2 torch==2.7.0 torchmetrics==1.7.1 lightning==2.5.1.post0 lightning-utilities==0.14.3 umap-learn==0.5.7 pyro-api==0.1.2 pyro-ppl==1.9.1 statsmodels==0.14.4 tqdm==4.67.1