From specimen to clinical decision. I design and operate genomics analysis that ends up in reports and services a hospital actually uses, and I measure how far those results can be trusted.
Chief Researcher, Diagnosis Division, Cytogen — CTC-based liquid biopsy, Seoul, Korea.
20+ years omics and clinical data · 20+ peer-reviewed papers · 7 institutions and companies · 4 degrees across computer science, life science, and molecular medicine
Portfolio · Blog · CV · Google Scholar · ORCID · LinkedIn · kakyung.kim@gmail.com
- 2026.11 — Oral presentation accepted at BIOINFO 2026 / GIW XXXV / ISCB-Asia (22 of 287 submissions): a reliability map for per-gene multiome RNA velocity parameters. A second abstract from the same harness was accepted as a poster.
- 2026 — Shipped a cell image review assistant in-house, trained on 1,531 cells drawn from reader decisions. Slide-level cross-validation had inflated agreement to kappa 0.910; re-splitting at the specimen level to block leakage brought it to 0.704, and the corrected figure is the one that shipped. Confidence-based triage cut the cells needing priority human confirmation by 70%.
- 2026 — Published an ADMET split audit on public data. Across 12 Therapeutics Data Commons tasks, a structure-disjoint split degraded all 12 (median drop 0.098); on screening metrics, top-10% hit rate fell on all 8 tasks tested and VDss went from 0.358 to 0.176 (p<0.001). Raising seeds from 5 to 15 narrowed the gaps enough to retract three of my own initial conclusions, and I published the correction. Write-up (Korean)
- 2026 — Fifth trend report in BRIC View (2026-T24, generative AI in biotech and drug discovery).
- Since 2026.06 — Two agent harnesses running in the open: econ-radar daily, paper-radar weekly.
| Year | Work | Venue | Role |
|---|---|---|---|
| 2019 | Whole-exome and whole-transcriptome sequencing of canine mammary gland tumors | Scientific Data 6(1):147 | First author |
| 2013 | miRTCat: a comprehensive map of human and mouse microRNA target sites | Bioinformatics 29(15):1898 | First author |
| 2018 | Palmitate and mmLDL cooperatively promote inflammatory responses in macrophages | PLoS One 13:e0193649 | Co-first author |
| 2020 | Cross-species oncogenic signatures of breast cancer in canine mammary tumors | Nature Communications 11:3616 | Co-author |
| 2022 | GWAS identifies TNFSF15 associated with childhood asthma | Allergy 77(1):218 | Co-author |
| 2021 | Genetic features of gastric mixed adenoneuroendocrine carcinomas | Journal of Pathology 253(1):94 | Co-author |
Full list on Google Scholar, ORCID, and the portfolio. Three manuscripts in preparation.
Automate the interpretation, but leave the evidence so a person can check it.
| Repository | What it does |
|---|---|
| econ-radar | Agent harness publishing a daily economic digest weighted toward pharma, bio, and AI. Collection, selection, summarization, and publishing split across agents. Daily brief plus weekly roundup, running since June 2026. |
| paper-radar | Weekly digest of bioinformatics, clinical ML, and drug-discovery AI papers, sorted into three readings: method, clinic, and industry. |
| paper-production-harness | Topic-agnostic multi-agent harness that runs a paper from analysis through to the talk. CC BY 4.0. |
| korea-agentic-hackathon-2026 | Drug candidate validation agent that catches claims a docking score cannot support. NVIDIA × FastCampus Korea Agentic AI Hackathon 2026. |
| pharmasignal-v0 | Drug safety signal agent over regulatory filings and trial registries. Inference is routed by data sensitivity, so restricted material never reaches a commercial API. Agent Forge AI Hackathon Seoul 2026. |
| chillmcp | MCP server built with FastMCP: 12 model-callable tools and 5 interactive prompts, MIT licensed. SKT AI Summit Hackathon 2025. |
Closed-source work includes VeriVar, a dual-mode somatic variant interpretation pipeline where an LLM mode and a rule-based mode run side by side and their disagreements serve as a verification signal, and AutoBioX, the multi-agent research harness behind the GIW 2026 abstracts.
- BIOINFO 2026 / GIW XXXV / ISCB-Asia — oral presentation, accepted 22 of 287 (Nov 2026, upcoming)
- PseudoCon 2026 — the AutoBioX multi-agent pipeline
- ABDD Summit 2026, Stanford — automated clinical reporting for oncology NGS: a dual-mode LLM pipeline (i-Talk and poster)
- Future of Medicine Symposium 2025 — single-cell QC standardization and an scRNA-seq × immunofluorescence pipeline (lightning talk)
- BRIC View — five invited trend reports (2017, 2018, 2025, 2026 ×2) on cloud genomics, allergy therapeutics, single-cell in drug discovery, bio big data, and generative AI
I write up what I built and where I went wrong, in Korean and English: kakyungkim.github.io.
Korea Clinical Datathon, Grand Prize (2019, sepsis prediction, KoNECT) · Cheiron ResearchThon, Creative Academic Impact Award (2025) · ABDD Summit, Best Poster (2024) · KSMCB Annual Meeting, Outstanding Poster (2011)
- Omics — multi-platform NGS (WES, WGS, RNA-seq, scRNA-seq, multiome, liquid biopsy), Nextflow, Seurat/Scanpy, AWS
- AI — PyTorch, pathology foundation model embeddings, LLM agent harnesses, RAG with citation grounding, MCP
- Clinical — Ph1–3 trial analysis, clinical report automation, GCLP/CAP/ISO quality systems
- Languages — Python, R, Bash, SQL
Ph.D. Molecular Medicine, Sungkyunkwan University (2011) · M.S. Life Science, Korea University (2004) · B.S. Computer Science, Dongguk University (2000) · B.S. Health & Environment and Business, Korea National Open University (2023)