Haron Xu

Biomedical Engineering · Medical Information Engineering · Class of 2027

I am an undergraduate researcher at Sichuan University, working on generative models for scientific discovery.

My research spans biological sequence design, spatial omics, and computational pathology. I am particularly interested in structured generation, multimodal representation learning, and post-training scientific foundation models with verifiable biological feedback.

I conduct research with Prof. Xiao Han at Sichuan University and Prof. Rex Ying at Yale University.

Email GitHub CV

Research Interests

Generative Models

A natural image emerging from noise through a generative denoising process

Spatial Omics

Gene-expression heatmap emerging through a diffusion denoising process

Pathology

Low-resolution H&E whole-slide image of COAD TENX111 tissue

COAD TENX111 tissue represented by GenBio PathFM PCA colors

COAD TENX111 tissue represented by GigaPath PCA colors

COAD TENX111 tissue represented by UNI PCA colors

RNA Design

Protein Generation

A protein backbone emerging through the RFdiffusion denoising process

My work asks how scientific structure and biological constraints should shape generative objectives, what cross-modal information these objectives encode, and how pretrained models can be adapted through feedback and verification. A longer account of these questions is available on the Research page.

News

Sep 14, 2026

New personal essay: On the Eve of an Intelligence Explosion is here.

Sep 10, 2026

New research blog: Domain-Specific Model Design as a Search Problem is here.

Jun 13, 2026

BioFlow was accepted to MICCAI 2026.

May 12, 2026

SpaMV was published in Nature Communications.

Selected Publication

MICCAI
BioFlow: A Biologically Valid Support-Preserving Flow for Histology-Conditioned Spatial Transcriptomics Prediction

Haoran Xu, Yang Liu, Wei Yuan, and 1 more author

In International Conference on Medical Image Computing and Computer-Assisted Intervention, 2026

Abs Bib Code

BioFlow is a support-preserving flow-matching framework for histology-conditioned spatial transcriptomics prediction. It preserves the non-negative support of gene-expression data while improving spatial correlation, training stability, and computational efficiency.
@inproceedings{xu2026bioflow, title = {{BioFlow}: A Biologically Valid Support-Preserving Flow for Histology-Conditioned Spatial Transcriptomics Prediction}, author = {Xu, Haoran and Liu, Yang and Yuan, Wei and Han, Xiao}, booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention}, year = {2026}, }

Nat Commun

Interpretable spatial multi-omics data integration and dimensionality reduction with SpaMV

Yang Liu, Kexin Ma, Haoran Xu, and 9 more authors

Nature Communications, 2026

Abs DOI Bib HTML Code

@article{liu2026spamv,
  title = {Interpretable spatial multi-omics data integration and dimensionality reduction with {SpaMV}},
  author = {Liu, Yang and Ma, Kexin and Xu, Haoran and Xu, Ke and Hu, Yunfei and Lin, Zhenhan and Lin, Jiangli and Han, Bo and Li, Shuaicheng and Lin, Zhixiang and Zhou, Xin Maizie and Zhang, Lu},
  journal = {Nature Communications},
  volume = {17},
  pages = {8213},
  year = {2026},
  doi = {10.1038/s41467-026-74718-1},
}

My Blog

Research Ideas behind the work

Longer reflections on generative modeling, biological AI, scientific constraints, and questions emerging from ongoing research.

Explore research writing Notes Things worth keeping

Short records of useful tools, technical workflows, experiments, and occasional thoughts beyond formal research projects.

Browse field notes