New Paper🌟. How can we accurately infer gene expressions from histology images? We present 𝐒𝐭𝐞𝐦🍀, a novel approach based on diffusion model. Glad to share that our work has been accepted to #ICLR2025 🎉. 📍Paper: 📍Code:

RT @YuchenZhu_ZYC: 🚀 New Paper at #icml2025🚀. Diffusion models have succeeded in handling data of various modalities, such as 🖼️images (co….

RT @KevRojas1499: 🌐 Diffusion models have succeeded on:. - Euclidean (images).- Discrete (language).- Riemannian (manifolds). But what if d….

Joint work with @yuchen4975, @MoleiTaoMath, Peng Qiu, @GeorgiaTech.

𝐒𝐭𝐞𝐦 accurately captures biological heterogeneity, which boosts its performance on downstream tasks such as identification of marker genes, unsupervised tissue structure annotation, etc.🔬

𝐒𝐭𝐞𝐦 achieves SOTA performance on data acquired from different species, organs, and health conditions. 🆕We also propose a new evaluation metric that measures the recovery of gene variation, which is largely overlooked in the current evaluation of gene inference tasks.

We propose a multimodal diffusion model conditioned on the visual knowledge distilled from multiple off-the-shelf pathology foundation models. By embedding gene expressions in the log-transformed space, our model accurately reconstructs the expression at its original scale🤩.

Spatial Transcriptomics bridges gene expressions and histology images, allowing inference of molecular-level information from easily accessible histology images. We formulate the inference problem as a generative modeling task, which enables us to tackle it with diffusion models.