Abstract:Inferring spatial transcriptomics (ST) from histology enables scalable histogenomic profiling, yet current methods are largely restricted to single-tissue models. This fragmentation fails to leverage biological principles shared across cancer types and hinders application to data-scarce scenarios. While pan-cancer training offers a solution, the resulting heterogeneity challenges monolithic architectures. To bridge this gap, we introduce MoLF (Mixture-of-Latent-Flow), a generative model for pan-cancer histogenomic prediction. MoLF leverages a conditional Flow Matching objective to map noise to the gene latent manifold, parameterized by a Mixture-of-Experts (MoE) velocity field. By dynamically routing inputs to specialized sub-networks, this architecture effectively decouples the optimization of diverse tissue patterns. Our experiments demonstrate that MoLF establishes a new state-of-the-art, consistently outperforming both specialized and foundation model baselines on pan-cancer benchmarks. Furthermore, MoLF exhibits zero-shot generalization to cross-species data, suggesting it captures fundamental, conserved histo-molecular mechanisms.
| Comments: | Accepted at Proceedings 43rd International Conference on Machine Learning, Seoul, South Korea |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2602.02282 [cs.LG] |
| (or arXiv:2602.02282v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2602.02282 arXiv-issued DOI via DataCite |
|
| Journal reference: | Proceedings 43rd International Conference on Machine Learning 2026 |
From: Susu Hu [view email]
[v1]
Mon, 2 Feb 2026 16:23:31 UTC (14,766 KB)
[v2]
Wed, 6 May 2026 14:12:08 UTC (14,765 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。