Yunsoo Kim's website

Yunsoo
Kim
Yunsoo
Kim
Yunsoo
Kim
Yunsoo
Kim

Yunsoo
Kim
Yunsoo
Kim
Yunsoo
Kim
Yunsoo
Kim

PhD candidate

A BRIEF SUMMARY ABOUT ME

I'm Yunsoo Kim, a PhD candidate at University College London
under the supervision of Professor Honghan Wu.
Previously, I was a professional (principal researcher) and NLP group leader at LG Chem, and a co-founder of Torrey Pines Co., Ltd. (microbiome science cosmetics company).

I received my MSc in Bioinformatics and Theoretical Systems Biology from Imperial College London, where I developed a machine learning method for -omics time series analysis with Professor Tim Ebbels,
and a metagenomics networks inference method with Dr. John Pinney and Dr. Virginia Fairclough.
I received my B.S. in Systems Biology with Honors (as well as concentration in Bioinformatics and Pre-Health Plan) from Case Western Reserve University,
where I made an assembly of Marama chloroplast genome with Professor Christopher Cullis.

My current work is focused on multimodal large language models and its applications in health informatics.
My research interest is Multimodal LLMs for histopathology and radiology, specifically in collating clinical expert knowledge such as eye gaze and knowledge base.

PLACES I LIVED IN

2023 ~
2017 ~ 2023
2015~2016
2008~2015
2007~2008
~ 2006

United Kingdom
South Korea
United Kingdom
United States
Canada
South Korea

RESEARCH INTEREST

Health

Histopathology

Radiology

Large Language Model

Multimodal

Human Expertise

HIGHLIGHTED WORKS

NEWS

[June 2025] IHC-LLMiner for PubMed trends analysis work accepted Oral presentation at European Congress of Pathology 2025.
[May 2025] 2 works accepted at ACL2025 as Findings. Look & Mark - Leveraging Radiologist Eye Fixations and Bounding boxes in Multimodal Large Language Models for Chest X-ray Report Generation and BioHopR - A Benchmark for Multi-Hop, Multi-Answer Reasoning in Biomedicine.
[April 2025] IHC-LLMiner demo (oral) and LLM instruction tuning for markdown table extraction (poster) accepted at HealTac2025
[April 2025] Poster presentation and Lightning talk at RECOMB2025 for the Real world histopathology reports LLM analysis work.
[February 2025] Organising 1 Workshop and 1 Tutorial at MICCAI2025.
For all the publications, please refer to google scholar .

PhD Thesis: Foundation Model in Medicine

Multimodal LLM for biomedical domain collating clinician's expertise.
I was an organiser of the tutorial at HealTac2024 and MICCAI2024.
I am currently organising the tutorial and workshop at MICCAI2025.
Thesis Proposal Presented at ACL2024 Student Research Workshop.

PUBLICATION

Enhancing VLMs in CXR analysis with Eye Gaze

A novel approach to enhance human-computer interaction in chest X-ray report generation
with radiologists’ eye gaze as well as abnormality grounding.
The work is accepted at ACL2025 as Findings.

BioHopR: Novel Benchmark for Medical LLM multi-hop reasoning

We address a major gap in current medical QA benchmarks by evaluating multi-hop, multi-answer reasoning in structured biomedical knowledge graphs.
The work is accepted at ACL2025 as Findings.

IHC-LLMiner: automatically extracting immunohistochemistry (IHC) marker-tumour profiles.

This study presents an automated pipeline, IHC-LLMiner,
for extracting IHC-tumour profiles from PubMed abstracts,
leveraging large language models. Our fine-tuned model outperformed GPT4O.
The work is under review.

Enhancing VLMs in CXR analysis with Eye Gaze

A novel approach to enhance human-computer interaction in chest X-ray analysis
with radiologists’ attention by incorporating eye gaze data alongside textual prompts for VLMs.
Oral Presentation at IJCAI2024 TAI4H workshop and Poster Presentation at MICCAI2024 main conference.

MedExQA: Novel Benchmark for Medical LLM

We address a major gap in current medical QA benchmarks
which is the absence of comprehensive assessments of LLMs’ ability to generate nuanced medical explanations.
This work also proposes a new medical model, MedPhi-2,
which outperformed medical LLMs based on Llama2-70B in generating explanations.
The work is presented at ACL2024 BioNLP workshop as a poster.

Chemical Language Understanding Expert

Working on language (pre-trained) models in chemistry, materials science, and synthetic biology patents and literature.
Working on knowledge graphs construction using NLP and GNN.
Developing a system called CLUE (chemical langauge understanding expert), which is an artificial reading system empowered by the fine-tuned models such as chemical entity recognition models.
Received 1st place in 2021 LG AI DX IDEAthon business applications sector. Published in ACL2023 Industry Track. 5 patents for this research.

PATENT

Protein Toxicity Prediction

Early Allergen and Toxin assessment system 2018
Developed a neural network to predict protein toxicity with protein amino acid sequence as input.
precision 2% higher and prediction speed 22 times faster than SVM (which was the SOTA).

Torrey Pines

Co-founder 2015 - present & CEO 2016 - 2017
Torrey Pines is a cosmetics company specialized in hair care products.
Introduced microbiome science to hair care cosmetics marketing.
Successfully exported to Vietnam and Cambodia in 2017.
Launched a new brand called "Bota nouveau" in Cambodia (2019) and Korea (2020).

Honors and
Other Experiences

Led UCL team for MICCAI2024 tutorial
The largest tutorial at MICCAI2024
More than 100 attendees.

Tutorial at MICCAI2024

Received the Best Award in LG for AI research
Given by LG Group Chairman
Due to COVID19, LG Chem CEO
handed out the award.

LG AWARD

Introduced microbiome science
to hair care cosmetics.
Actively used microbiome science for marketing.

First brand

I received FACT Graduate Scholarship which supports outstanding PhD students
in innovative, quantitative, and interdisciplinary research.

FACT Graduate Scholarship 2024

Received DX Frontier 1st place in applications sector
Given by LG Science Park
Was part of LG AI IDEAthon
Handed out the award in Metaverse.

DX Frontier 1st Place

Mentoring of 4 researchers at LG in 2020
One mentee received best award from the mentoring.
He continued working on it and received LG best practice in 2021.