GeneFlow
A gene expression prediction framework for spatial transcriptomics (ST) that combines flow-matching generative modeling with structured multimodal fusion to infer the expression of large gene panels directly from histology images without complex generative pipelines.
Timeline (last 90 days)
GeneFlow is a framework that unifies flow-matching generative modeling with structured multimodal fusion for predicting gene expression from tissue images in spatial transcriptomics.
GeneFlow uses a histology tokenizer and a gene tokenizer to encode tissue morphology information and cross-gene interactions within spatial spots, respectively.
GeneFlow directly predicts spatial gene expression rather than an intermediate velocity field, and does not run the full generative trajectory at inference.
GeneFlow achieved state-of-the-art performance across multiple spatial transcriptomics benchmarks.
According to the paper, GeneFlow improves performance by up to 56.8% on the task of predicting 1000 genes.
GeneFlow reduces computational cost by several orders of magnitude.
The spot expression profiles generated by GeneFlow are biologically meaningful, not merely numerically fitted.
GeneFlow's code is publicly available on GitHub at https://github.com/ZZhangsm/GeneFlow.
GeneFlow is expected to promote the widespread application of spatial transcriptomics in clinical and research settings.
All Facts (9)
GeneFlow is a framework that unifies flow-matching generative modeling with structured multimodal fusion for predicting gene expression from tissue images in spatial transcriptomics.
50%UnverifiedGeneFlow uses a histology tokenizer and a gene tokenizer to encode tissue morphology information and cross-gene interactions within spatial spots, respectively.
50%UnverifiedGeneFlow directly predicts spatial gene expression rather than an intermediate velocity field, and does not run the full generative trajectory at inference.
50%UnverifiedGeneFlow achieved state-of-the-art performance across multiple spatial transcriptomics benchmarks.
50%UnverifiedAccording to the paper, GeneFlow improves performance by up to 56.8% on the task of predicting 1000 genes.
50%UnverifiedGeneFlow reduces computational cost by several orders of magnitude.
50%UnverifiedThe spot expression profiles generated by GeneFlow are biologically meaningful, not merely numerically fitted.
50%UnverifiedGeneFlow's code is publicly available on GitHub at https://github.com/ZZhangsm/GeneFlow.
50%UnverifiedGeneFlow is expected to promote the widespread application of spatial transcriptomics in clinical and research settings.
50%