lazyslide_models.feature_prediction.Path2Space#
- class Path2Space(model_path=None, token=None)#
Bases:
FeaturePredictionModelpath2space GitHub Paper Params: 1.76B Apache-2.0 [Shulman et al., 2026] Predict spatial gene expression from CTransPath tile features. Path2Space gene-expression predictor.
The model consumes raw 768-dimensional CTransPath tile features and returns one NumPy array per gene. Gene names and output ordering are read from the gene list published alongside the exported ensemble.
Upstream extracts those features from Macenko stain-normalized tiles (the companion pipeline’s frozen normalizer). Normalize tiles the same way, e.g. with a
transformpassed to feature extraction.- predict(features)#
Predict expression for an
[n_tiles, 768]feature matrix.