Portrait of Soyon Park

Soyon Park

[sʌ.jʌn pak]

PhD Student, Biomedical Informatics, Harvard Medical School Bioinformatics and Integrative Genomics (BIG) Track

I am a first-year PhD student in Biomedical Informatics at Harvard Medical School, in the Bioinformatics and Integrative Genomics (BIG) track. I develop machine learning methods for single-cell biology, with a focus on causal learning and generative models for predicting how cells respond to genetic and chemical perturbations. I am also interested in spatial omics, linking tissue morphology to spatially resolved gene expression.

Before Harvard, I received my MS in Computer Science & Engineering from Korea University, where I worked in the DMIS Lab with Prof. Jaewoo Kang, and my BS in Life Sciences and Bioinformatics from Ewha Womans University. I have also worked with Prof. Maria Brbić at EPFL and Prof. Luca Pinello at MGH / Harvard Medical School.

Bioinformatics
Single-cell omics, spatial transcriptomics, multi-omics, perturbation modeling
Causal learning
Causal discovery, counterfactual reasoning, disentangled representations
Deep learning
Generative models, graph neural networks, foundation models

News

Publications

* equal contribution

Full list on Google Scholar.

Ongoing Research

EPFL · MLBio Lab
Foundation model for single-cell phenotype modeling

Predicts gene expression responses to perturbations and, in reverse, infers the perturbation from post-perturbation profiles, across genetic, chemical, and developmental settings.

MGH / HMS · Pinello Lab
Coral: probabilistic modeling for single-cell multiomics

A probabilistic framework with a heterogeneous graph transformer that learns gene–peak relationships across scRNA-seq and scATAC-seq.

ISMB/ECCB 2025 poster
SPARTHA: spatial gene expression from histology

A multimodal conditional VAE that generates spatial gene expression from tissue morphology while disentangling batch effects.

Korea University · DMIS Lab
Multi-cancer biomarkers from plasma MS/MS

Transformer-based, ID-free representation learning on peptide DIA MS/MS spectra to find cancer-specific embeddings and classify ten cancer types.

Experience

Education

Honors & Scholarships