%matplotlib inline

Available pretrain models

We pretrained models on preprocessed high-quality spatial transcriptomics data, and realised them for transfer learning for new spatial data, so that people can do segmentation from scratch without providing coarse pre-trained labels. Currently we cover pre-trained models from multiple tissues and spatial technologies.

Import packages & data

import random
import numpy as np
import pandas as pd
import tifffile as tiff
import matplotlib.pyplot as plt 

import Bering as br

Pre-train model for ISS hippocampus

This model is pre-trained on ISS (pciSeq) hippocampus data. Available labels contains subtypes of neuron and non-neurons.

bg = br.datasets.model_iss_ca1_qian()

# parameters of node classification model
bg.trainer_node.model.parameters()

# parameters of edge prediction model
bg.trainer_edge.model.parameters()

# available labels for node classification
bg.labels

# number of input node features
bg.n_node_features

Other available models

  • Cortex (MERFISH): br.datasets.model_iss_ca1_qian()

  • Lung cancer (Cosmx): br.datasets.model_cosmx_nsclc_he()

  • Breast cancer (Xenium): br.datasets.model_xenium_dcis_janesick()

  • Hippocampus (pciSeq): br.datasets.model_iss_ca1_qian()

More pretrained models will be added.