Introduction

A few years ago, I lost a family member to blood cancer. I wanted to do something, anything, and searched everywhere for answers, but there was nothing I could actually do. Watching them have no choice but to rely on standard care alone was devastating. The sense that there had to be something more, and that I couldn’t do any of it, stayed with me for a long time.

In time, I came across the case of Sid Sijbrandij. GitLab co-founder Sid Sijbrandij examined his rare-cancer data in depth and published the process of reviewing treatment candidates with specialists. In that case, I saw the outline of the “something I could do” that I had never managed to find back then.

Genomic analysis is hard, and turning it into something that actually helps treatment is, honestly, far more likely to fail than to succeed. It may, in the end, never amount to real help for anyone. But it still isn’t something we can simply choose not to do. So we decided to study Sid’s case together, write it up, and share it openly. HomeGenomics is that record. Korea has no shortage of founders and developers who are good at solving hard problems, and our goal is to help them grow a little more comfortable with this field.

This handbook isn’t put together by me alone: it’s compiled by a study group of people working in bioinformatics, pharmacy, and computer science. We bring different backgrounds to the same material and review the code and interpretations together.

Who this handbook is for

HomeGenomics is a handbook that helps developers analyze their own and their family’s genomic data themselves. You don’t need a biology degree: if you can write code, you can start.

• People who received raw genetic test data and want to look beyond the summary report
• People who want to understand and prepare for conditions that run in the family, using data
• Programmers who are new to bioinformatics
• Anyone curious about how precision medicine actually plays out, through a real case

How you learn on this site

Now that AI agents can write code and run analysis tools, memorizing every command and option matters less than being able to specify what to analyze and judge the result.

HomeGenomics focuses on three responsibilities that cannot be delegated to an agent.

0. Understand the underlying principle: understand how input data becomes a result through particular assumptions and calculations.
1. Decide what to examine: specify which data to compare, which signal to look for, and which outputs to produce.
2. Interpret the result: distinguish what a table or figure supports, what it cannot establish, and which question should come next.

Step-by-step agent prompts delegate environment setup, command execution, and Python coding. A prompt is not an order for the desired answer. It is an analysis specification that defines the inputs, comparison, outputs, and validation checks.

The reader’s role is not to copy code. It is to inspect what the agent produces, question assumptions and outliers, and decide where the analysis should go next.

Principles we follow

HomeGenomics follows these standards when creating and revising content.

1. Prefer public and synthetic data: in real cases, we share analysis questions and methods rather than personal details. If personal or family data is necessary, we first define its purpose, storage location, access, and retention period, and check what leaves the local environment through files, prompts, and logs.
2. Treat AI output as something to review: an AI’s statement is not evidence. We check whether the input data, calculation, actual artifacts, and sources agree. When an agent should not decide on its own, it must present the available choices and stop.
3. Write only what the evidence supports: we distinguish observations, a participant’s account, company claims, media reports, and HomeGenomics’s interpretation. We do not generalize from one person or experiment, or treat finding a candidate as proof of an effect.

Important precautions

HomeGenomics is a handbook for learning how to analyze and interpret genomic and biological data, primarily through public and synthetic datasets. Its tables, figures, and candidate lists are not diagnostic or treatment instructions. Do not start, stop, or change medication, testing, or treatment based solely on this site or an AI agent’s output. Decisions about diagnosis and treatment in people or animals, and about manufacturing or administering medical products, require qualified professionals and institutions that can evaluate validated tests, clinical context, benefits, and risks together.

Personal and family genomic data is sensitive information that cannot be changed like a password. Practice first with the public or synthetic data specified by this site. Before handling personal data, decide where it will be stored and who it may be shared with.

Next: The case of Sid Sijbrandij

---

Questions, feedback, or corrections: [email protected]