arXiv：cs.LG（机器学习，全量分类）· Alistair Wilkinson, Christopher J. Tape, Smita Krishnaswamy·· 1 天前AI 评分32
Pheno-GS：面向 Phenoscape 规模的最短路径 Sinkhorn 算法
Pheno-GS: Phenoscape-scale Geodesic Sinkhorn
AI 导读
Pheno-GS 可在噪声、不平衡、大规模场景下计算准确且可扩展的测地线传输距离，对 500 个分布的计算速度比 Geodesic Sinkhorn 快 200 倍以上。该方法通过图连通性正则化、KL 边缘惩罚的不平衡 OT 公式以及批量矩阵算法三个组件实现，已在合成基准和 CyTOF 扰动数据集上验证，论文被 IEEE MLSP 2026 接收。
正文
Abstract:High-throughput single-cell data is now collected across large patient cohorts. Understanding patient-level heterogeneity from cellular-level data motivates phenoscaping: embedding each single-cell distribution as a "datapoint," with distances given by optimal transport (OT). Computing geometry-aware OT at this scale, between all pairs of patient datasets, remains an open challenge, since existing methods either rely on Euclidean ground metrics that distort manifold structure or fail under sparse, unevenly sampled, or large-scale data. We present \textbf{Pheno-GS} (Phenoscape-scale Geodesic Sinkhorn), which computes accurate, scalable geodesic transport distances under noisy, unbalanced, large-scale settings via three components: ($1$) graph connectivity regularization for well-defined geodesics on sparse/disconnected manifolds; ($2$) an unbalanced OT formulation via KL marginal penalties; and ($3$) a batched matrix algorithm computing all pairwise distances in one heat diffusion (over $200 \times$ faster than Geodesic Sinkhorn for $500$ distributions). We validate Pheno-GS on synthetic benchmarks and a CyTOF perturbation dataset.
| Comments: | Camera-ready version accepted at IEEE MLSP 2026; notation corrections to mathematical typesetting |
| Subjects: | Machine Learning (cs.LG); Quantitative Methods (q-bio.QM) |
| Cite as: | arXiv:2609.27633 [cs.LG] |
| (or arXiv:2609.27633v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.27633 arXiv-issued DOI via DataCite |
Submission history
From: Alistair Wilkinson [view email]
[v1]
Wed, 23 Sep 2026 09:54:58 UTC (23,051 KB)
[v2]
Thu, 1 Oct 2026 03:47:35 UTC (23,051 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org