Spatial Transcriptomics as Images for Large-Scale Pretraining
Researchers have developed a novel method to represent spatial transcriptomics data as images for large-scale pretraining. This approach treats tissue sections as croppable image patches, allowing for a significant increase in training samples while preserving spatial context. Experiments demonstrate that this image-like dataset construction consistently improves downstream performance compared to existing pretraining schemes. AI
IMPACT This new method could enable more effective large-scale pretraining of AI models on biological data, potentially accelerating discoveries in pathology and clinical studies.