Model Zoo#

The repository contains wrappers for three spatial proteomics foundation models, a ResNet and a linear-regression baseline, three H&E-to-multiplex virtual staining models, three cell instance segmentation models, and two specialized cell-typing tools. Each is instantiated from its model config.

Model

Model config

Supported tasks

VirTues

configs/models/virtues.yaml

In-cohort cell phenotyping, tissue-level tasks, and virtual staining

KRONOS

configs/models/kronos.yaml

In-cohort cell phenotyping and tissue-level tasks

Eva

configs/models/eva.yaml

MPX-to-MPX virtual staining only

ResNet

configs/models/resnet.yaml

Tissue-level tasks only

Linear Inpainter

configs/models/linear_inpainter.yaml

MPX-to-MPX virtual staining only

ROSIE

configs/models/rosie.yaml

H&E-to-MPX virtual staining only

HistoPlexer

configs/models/histoplexer.yaml

H&E-to-MPX virtual staining only

GigaTIME

configs/models/gigatime.yaml

H&E-to-MPX virtual staining only

Cellpose

configs/models/cellpose.yaml

Cross-cohort cell instance segmentation only

InstanSeg

configs/models/instanseg.yaml

Cross-cohort cell instance segmentation only

StarDist

configs/models/stardist.yaml

Cross-cohort cell instance segmentation only

astir

configs/models/astir.yaml

Cell phenotyping only, via a separate pipeline

MAPS

configs/models/maps.yaml

Cross-cohort cell-type annotation only, via a separate pipeline

Set the model-specific paths, such as checkpoints and marker embeddings, in each model config before running; see Configuration.

astir and MAPS#

astir and MAPS are not wrapped in SporaModelWrapper and run with their own pipelines. The astir pipeline always loads configs/models/astir.yaml. Its benchmark configs, which add marker lists per cell type, live in configs/benchmarks/astir/:

python -m spora_bench.benchmarks.run_astir \
  datasets_config=<path-to-datasets-config> \
  benchmarks_config=configs/benchmarks/astir/<dataset>.yaml

MAPS is trained on one set of datasets and evaluated on another:

python -m spora_bench.benchmarks.run_maps \
  model_config=configs/models/maps.yaml \
  train_datasets_config=<path-to-training-datasets-configs> \
  test_datasets_config=<path-to-test-datasets-configs> \
  benchmarks_config=<path-to-cell-annotation-benchmarks-configs>