Hyeongju Lee 이형주

Annotating genomic variants
against public references

The system links a person's genomic variants (VCF, chip arrays) to public references and reports disease and trait risk, population frequencies, and related drugs and herbal medicines. The analysis algorithm was designed at the company with my team lead; I implemented and deployed the pipeline, backend and frontend. It was an international collaboration with Kyung Hee University's College of Korean Medicine and a Canadian bio research lab.

Period
Aug 2022 – Sep 2023Vision Inside · internal project
Role
Pipeline, backend and frontendIncluding Docker deployment
Design
Algorithm with the team leadUI design by the Canadian lab's designer
Outcome
Deployed, internal IPs onlyNever launched as a public service
SQL join annotation4stagesReferences attached in turn by position and alleles, or rsID
1000 Genomes26populationsAllele frequencies compared
DeploymentDocker· HarborKorea IDC production server, private registry set up myself

Architecture

The flow from input to the report API. Click a node to see the path through it and who did what.

Implemented by me Shared with others Others Input and external data
  1. 01Input
    • Input
      Raw sequencing data (FASTQ), the input to the variant-calling pipeline.
    • Input
      A VCF file or a genotyping-array (chip) result file uploaded by the customer.
    • Data
      dbSNP (GRCh37, GRCh38), 1000 Genomes, the GWAS Catalog and UniProt.
  2. 02Processing
    • What I did
      Implemented the FASTQ → BWA → samtools → Picard → GATK → VCF pipeline.
      Others
      The pipeline algorithm was designed at the company together with my team lead.
      Stack
      • BWA
      • samtools
      • Picard
      • GATK
    • What I did
      Read customer VCF and chip (array) input with PyVCF and shaped it into the input table used for annotation. Chip input finds its position key in dbSNP by rsID.
      Stack
      • PyVCF
      • Python
  3. 03Storage
    • What I did
      Built dbSNP, 1000 Genomes, the GWAS Catalog and UniProt into MySQL. Split the files by chromosome (sed, split) and loaded them in pandas chunks through SQLAlchemy.
      Stack
      • MySQL
      • pandas
      • SQLAlchemy
      • sed · split
  4. 04Annotation
    • What I did
      A four-stage SQL join: dbSNP and 1000 Genomes on a key of chromosome, position, REF and ALT, the GWAS Catalog on rsID and risk allele, and UniProt on variant ID and allele. A rule-based risk score from GWAS odds ratios and betas, allele frequencies compared across the 26 populations of 1000 Genomes, and drug and herbal-medicine information per disease (MedlinePlus, MeSH, ATC, PharmDB-K, OASIS).
      Note
      The LDA and 1D-CNN models linking diseases to herbal medicines stayed prototype experiments.
      Stack
      • SQL
      • MySQL
      • Python
  5. 05Serving
    • Others
      The screens were designed by a designer at the Canadian bio research lab.
      What I did
      Worked out the requirements with the designer and handled technical communication between the Korean and Canadian teams.
    • What I did
      Served the reports through a FastAPI JSON API and also built the user-facing screens (frontend). Later I removed the web screens from the service and turned it into a JSON API for the frontend.
      Stack
      • FastAPI
      • JSON API
    • What I did
      Set up a Harbor private registry myself on an Ubuntu production server at Korea IDC (certificate, domain, Let's Encrypt) and deployed with Docker images. The production server was kept reachable from internal IPs only.
      Stack
      • Docker
      • Harbor
      • Ubuntu
      • Let's Encrypt
ImprovedWeb screens rendered inside the server → a separate JSON API for the frontend
ResultDeployed to a production server and run for internal IPs onlyThe service code was cleaned up and published

References

The public data used for annotation.

1000 Genomes
Allele frequencies in 26 populations
dbSNP
GRCh37 · GRCh38, IDs and positions
GWAS Catalog
Disease and trait associations, odds ratios
UniProt
Protein variants and clinical significance

Loading and annotation

Too large to load in one go, so it went in piece by piece.

The dbSNP and 1000 Genomes source VCFs were too large to load in one go, so I split the files by chromosome (sed, split) and loaded them a little at a time, in pandas chunks through SQLAlchemy.

Annotation is a four-stage SQL join that attaches the reference tables one after another: dbSNP and 1000 Genomes on a key of chromosome, position, REF and ALT, the GWAS Catalog on rsID and risk allele, and UniProt on variant ID and allele. Chip input finds its position key in dbSNP by rsID.

Scope

What was designed together and what I implemented, listed separately.

What I implemented and deployed

  • Implementing the variant-calling pipeline (FASTQ → BWA → samtools → Picard → GATK → VCF)
  • Handling of customer VCF and genotyping-array input (PyVCF)
  • Loading public references and four-stage SQL annotation
  • Disease and trait risk scores from GWAS odds ratios, population frequencies, and drug and herbal-medicine information per disease
  • The FastAPI JSON API and the report frontend (user-facing screens)
  • Docker deployment, a Harbor private registry, domain and HTTPS
  • Technical communication between the Korean and Canadian teams, and aligning requirements with the Canadian lab's designer

Done together or by others

  • With the team lead · designing the pipeline algorithm
  • The Canadian lab's designer · UI design
  • Kyung Hee University's College of Korean Medicine · a Canadian bio research lab · partners in the international collaboration

Deployment and limits

What ran in production, and what stayed an experiment.

  1. Deployed with Harbor and Docker

    How
    I set up a Harbor private registry myself on an Ubuntu production server at Korea IDC (certificate, domain, Let's Encrypt), pushed the images there and deployed with Docker.
    Harbor
    Deployed to production through the Harbor registry I built, and used the same registry to set up a Kubernetes test environment with the team lead.
    Cleanup
    I removed the web screens and turned the service into a JSON API for the frontend.
  2. Limits

    Score
    Risk is a rule-based score from GWAS odds ratios and betas, not a machine-learning prediction.
    Models
    The LDA and 1D-CNN models linking diseases to herbal medicines stayed prototype experiments.
    Access
    After deployment to the production server it stayed reachable only from internal IPs. It was never launched as a public service.
  • Python
  • SQL
  • FastAPI
  • MySQL
  • pandas
  • SQLAlchemy
  • PyVCF
  • BWA
  • samtools
  • Picard
  • GATK
  • Docker
  • Harbor
  • Ubuntu
Contact

For roles and collaboration, please write by email.

I am looking for roles in data engineering, backend development and vision data pipelines.

Finishing my master's in March 2027