# HDTree: Generative Modeling of Cellular Hierarchies for Robust Lineage Inference

Source: [HDTree: Generative Modeling of Cellular Hierarchies for Robust Lineage Inference](<https://www.sota2.com/research/paper/68fa5f821e209131ce9ea80a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-20T05:00:20.896Z

Authors: Zelin Zang, WenZhe Li, Yongjie Xu, Chang Yu, Changxi Chi, Jingbo Zhou, Zhen Lei, Stan Z. Li

Published: 2025-06-29T15:19:13.000Z

## Abstract

In single-cell research, tracing and analyzing high-throughput single-cell differentiation trajectories is crucial for understanding biological processes. Key to this is the robust modeling of hierarchical structures that govern cellular development. Traditional methods face limitations in computational cost, performance, and stability. VAE-based approaches have made strides but still require branch-specific network modules, limiting their scalability and stability, while often suffering from posterior collapse. To overcome these challenges, we introduce HDTree, a generative modeling framework designed for robust lineage inference. HDTree captures tree relationships within a hierarchical latent space using a unified hierarchical codebook and employs a quantized diffusion process to model continuous cell state transitions. By aligning the generative process with the Waddington landscape, this method not only improves stability and scalability but also enhances the biological plausibility of inferred lineages. HDTree's effectiveness is demonstrated through comparisons on both general-purpose and single-cell datasets, where it outperforms existing methods in lineage inference accuracy, reconstruction quality, and hierarchical consistency. These contributions enable accurate and efficient modeling of cellular differentiation paths, offering reliable insights for biological discovery.\\footnote{Code is available at https://github.com/zangzelin/code\\\_HDTree\\\_icml.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Cellular Lineage Inference on Limb (cell lineage)](<https://www.sota2.com/research/sota/cellular-lineage-inference-on-limb-cell-lineage.md>): Cellular Lineage Inference; Limb (cell lineage); DP: 41.

- [Cellular Lineage Inference on LHCO cell lineage](<https://www.sota2.com/research/sota/cellular-lineage-inference-on-lhco-cell-lineage.md>): Cellular Lineage Inference; LHCO cell lineage; DP: 42.7.

- [Cellular Lineage Inference on Weinreb (cell lineage)](<https://www.sota2.com/research/sota/cellular-lineage-inference-on-weinreb-cell-lineage.md>): Cellular Lineage Inference; Weinreb (cell lineage); DP: 63.3.

- [Hierarchical Generative Modeling on MNIST](<https://www.sota2.com/research/sota/hierarchical-generative-modeling-on-mnist.md>): Hierarchical Generative Modeling; MNIST; DP: 92.7.

- [Hierarchical Generative Modeling on Fashion-Mnist](<https://www.sota2.com/research/sota/hierarchical-generative-modeling-on-fashion-mnist.md>): Hierarchical Generative Modeling; Fashion MNIST; DP: 57.4.

- [Hierarchical Generative Modeling on 20 Newsgroups text](<https://www.sota2.com/research/sota/hierarchical-generative-modeling-on-20-newsgroups-text.md>): Hierarchical Generative Modeling; 20 Newsgroups text; DP: 23.7.

- [Hierarchical Generative Modeling on Cifar10 32x32 (50k samples)](<https://www.sota2.com/research/sota/hierarchical-generative-modeling-on-cifar10-32x32-50k-samples.md>): Hierarchical Generative Modeling; Cifar10 32x32 (50k samples); DP: 44.2.

- [Cellular Lineage Inference on ECL cell lineage 838k points celltype:10](<https://www.sota2.com/research/sota/cellular-lineage-inference-on-ecl-cell-lineage-838k-points-celltype-10.md>): Cellular Lineage Inference; ECL cell lineage 838k points celltype:10; DP: 69.

- [Lineage Inference on LineageVAE Day 2](<https://www.sota2.com/research/sota/lineage-inference-on-lineagevae-day-2.md>): Lineage Inference; LineageVAE Day 2; Ratio of Observed Time Points: 23.2.

- [Lineage Inference on LineageVAE Day 4](<https://www.sota2.com/research/sota/lineage-inference-on-lineagevae-day-4.md>): Lineage Inference; LineageVAE Day 4; Ratio of Observed Time Points: 38.4.

- [Lineage Inference on LineageVAE Day 6](<https://www.sota2.com/research/sota/lineage-inference-on-lineagevae-day-6.md>): Lineage Inference; LineageVAE Day 6; Ratio of Observed Time Points: 62.

- [Lineage Inference on C. elegans 300–500](<https://www.sota2.com/research/sota/lineage-inference-on-c-elegans-300-500.md>): Lineage Inference; C. elegans 300–500; Ratio of Observed Time Points: 45.8.

- [Lineage Inference on C. elegans 500–750](<https://www.sota2.com/research/sota/lineage-inference-on-c-elegans-500-750.md>): Lineage Inference; C. elegans 500–750; Ratio Observed Time Points: 66.3.

- [Lineage Inference on C. elegans 100–300](<https://www.sota2.com/research/sota/lineage-inference-on-c-elegans-100-300.md>): Lineage Inference; C. elegans 100–300; Ratio of Observed Time Points: 15.2.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2506.23287>)

[PDF](<https://arxiv.org/pdf/2506.23287.pdf>)
