Read what usually happens.
Books, papers, and guidelines organize mechanisms and population evidence. They may stratify by known variables, but they do not instantiate a model of this person’s measured biological state.
Computational human biology Independent research
Toward models of individual biological response.
VHL’s long-term research program asks whether measurements from a person can become a computational representation of biological state and eventually support estimates of response to defined perturbations. Today’s work focuses on RNA methods, RNA-language representations, and leakage-controlled benchmarks.
Observe 01 Measured biology
Represent 02 Individual state
Perturb 03 Interventions + disease
Test 04 Outcomes + uncertainty
Definition
A model of biological state and response—not a visual replica.
In VHL’s research, a virtual human is a task-specific computational model conditioned on measurements from an individual. It asks how a measured state might change after a defined perturbation, and reports uncertainty and limits. It is not a complete digital copy or a clinical service available today.
Long-term research questions
What could become testable if an individual model were reliable?
These are research directions, not capabilities or services VHL offers today.
01
Virtual trials
Could compare responses before prospective testing
Use modeled cohorts to explore response heterogeneity and prioritize trial hypotheses before prospective human testing. Simulation cannot establish safety or efficacy.
02
Treatment response
Could model response to candidate medicines
Any eventual clinical use would require task-specific validation and independent clinician review of applicability, evidence, uncertainty, interactions, and safety.
03
Everyday exposures
Could study nutritional and behavioral exposures
Study response to nutrition, supplements, sleep, exercise, and other defined exposures across time rather than promise a personal recommendation.
04
Disease trajectories
Could estimate plausible disease trajectories
Use repeated measurements to estimate several plausible paths of progression, recovery, or relapse—with time horizons and uncertainty made explicit.
Why now
From reference knowledge to state-conditioned estimates.
Learned models offer a new interface to complex information. Biology can use that interface only when predictions are falsifiable, calibrated, and tested against new measurements.
Find what has been observed.
Search and databases make evidence available on demand. Retrieval improves access, but it does not itself produce an estimate conditioned on this person’s measured biological state.
Ask what may happen in this state.
A learned model can encode patterns from population data in parameters and condition an estimate on an individual’s measured state. It does not escape prior evidence; it reorganizes that evidence into a testable conditional estimate.
Current work / RNA
Current work begins with RNA.
RNA abundance provides a time- and tissue-dependent measurement of gene expression. VHL’s current projects ask how to resolve that signal, represent it for models, and test whether results transfer. These questions are relevant to the broader virtual-human research program without treating an RNA profile as the whole person.
01
Resolve / PBISC
PBMC bulk-to-single-cell generation
Test whether accessible bulk PBMC RNA can support a useful cell-resolved representation without presenting generated pseudo-cells as measurements.
02
Represent / RNA-LLM
Transcriptomes as model input
Encode single-cell, pseudobulk, or bulk RNA profiles into a learned representation for transcriptome-conditioned biological reasoning.
03
Test / Evaluation
Leakage-controlled model benchmarks
Hold out studies and donors so performance reflects transfer to new biology rather than recognition of repeated cohorts or individuals.
Manuscripts
Two manuscripts are under review.
Full PDFs are not posted publicly while patent filing and peer review are in progress. Researchers may request a private copy by email; each listing includes a blurred preview.
Latest essay
The Age of Trusting Parameters
Seung Jun Lee examines where individual-response models could enter drug development, why group averages leave a granularity problem, and what evidence must come first.
People
A small independent group across medicine and computational biology.
Affiliations identify individuals and do not imply institutional sponsorship, endorsement, or responsibility.
Active members
Seung Jun Lee
Seongjun Yang
Gyubok Lee
Pre-active members
Seung Hyun Hwang
Seungwoo Ryu
Advisors
Prof. Jung-Sik Song
Prof. Eun-Ju Lee
Contact
Discuss virtual-human research, a manuscript, or a collaboration.
Introduce your research context, affiliation, and the project or manuscript you want to discuss. Do not send patient information, controlled-access data, confidential material, credentials, or unpublished proprietary data by email.
junidude14@gmail.com