Hist2Cell: Deciphering fine-grained cellular architectures from histology images

Area of Science:

• Computational biology
• Genomics
• Pathology

Background:

• Histology images are cost-effective for predicting cellular phenotypes via spatial transcriptomics.
• Current methods lack accuracy in individual gene expression and fine-grained cell type prediction.

Purpose of the Study:

• To introduce Hist2Cell, a novel framework for resolving fine-grained transcriptional cell types directly from histology images.
• To enhance the accuracy and applicability of spatial transcriptomics analysis in cancer research.

Main Methods:

• Developed Hist2Cell, a vision graph-transformer framework.
• Trained the model on human lung and breast cancer datasets.
• Validated generalization on The Cancer Genome Atlas (TCGA) cohorts.

Main Results:

• Hist2Cell accurately predicts cell-type abundance (Pearson correlation > 0.80).
• The framework captures cellular colocalization and generalizes across large datasets without re-training.
• Identified relationships between tissue microenvironments, cell types, and patient mortality.

Conclusions:

• Hist2Cell enables cost-efficient, large-scale spatial biology studies.
• The framework facilitates precise cancer prognosis by analyzing tissue microenvironments.
• Hist2Cell advances the integration of histology imaging and transcriptomics for biomedical research.