Boyce Data Science · free tool
From Clinical Data to Omics
A map for analysts and statisticians who already work with clinical data and have been asked to run or lead a genomics or multiomic project.
Where you are now
Why I built this
Some years into a career built on registries, electronic health record (EHR) extracts, and study protocols, I was handed a project that produced sequencing files. I had no training in molecular biology. My first reaction was that I had been put in charge of something I could not read.
What I found instead was that most of my questions still applied. Where did these samples come from, and under what consent. Which version of the reference was used, and can I get that in writing. Who touched the file between the instrument and the table I am looking at. How many comparisons are we running, and against what. Those are provenance, governance, and inference questions, and they are the same questions I had been asking about claims extracts and chart abstractions for years. The vocabulary was new, the reasoning was not.
The gaps were real, and they were specific: file formats I had never opened, a compute environment I did not know how to log into, workflow software that looked nothing like a SAS program, and biology that gave the numbers their meaning. This tool is the map I wanted then, written for the person who can defend a cohort definition to a data governance committee and has never seen a FASTQ file.
Biology knowledge stays necessary, and this tool does not replace it. The goal here is to lead the project competently, ask the questions that keep it defensible, and read the output with the same skepticism you bring to a propensity score model. A bioinformatician on the team is not optional, and nothing below is a substitute for one.
Rate where you stand
Use the scale on each line. The first group is credit you already hold; the second is terrain that is new to most people coming from clinical data work. Your answers shape the recommended sequence in stage three and travel into the roadmap in stage five.
0 not yet · 1 some exposure · 2 comfortable working alone · 3 I teach this to others
Credit you already hold. These transfer into a genomics project with very little translation.
Terrain that is new. These are the skills the curriculum in stage three is built around.
Rate a few lines above and a reading of your profile appears here.
Name the project
What you already do, written a different way
Pick a task you perform today. The tool shows the same job on the command line and in R or Python, using one invented registry extract throughout, with a note on what changes when the file grows past the size a spreadsheet will open.
The running example
Everything below operates on ridgeline_extract_2026-03-01.csv, a comma delimited export from an invented study called the Ridgeline Registry. It holds one row per visit, with the columns person_id, site, enroll_date, birth_year, sex, dx_code, dx_date, visit_id, visit_date, value_num. A second file, codeset_seizure_dx.csv, holds one diagnosis code per row in a column named dx_code. The registry, the sites, and the codes are invented for this tool and correspond to no real organization or patient group.
Choose a task
The staged curriculum
The sequence below runs in the order a clinical data person can absorb it, each stage supported by the one before. Every stage opens with what changes for you once you have it, then what to learn, then freely available places to learn it. The flag on each stage comes from your ratings in stage one; change a rating and the flags move.
A caution about the middle of this list. Many programs run their sequencing analysis inside an established pipeline rather than assembling one command at a time, so recognizing what a pipeline is doing and where its configuration is recorded will take you further on most projects than writing one from scratch. Learn to read Snakemake or Nextflow before you set out to author either.
Project readiness check-up
This checks the project and the practices around it: where files live, how they are named, what is recorded, and who could reproduce the result. It examines your setup rather than any patient data, and it is not a clinical assessment of anything. Answer for the project as it stands today, not as you intend it to be.
Answer the items above and a summary of the findings appears here.
Prompts for the work
A language model is useful on this kind of project in a specific way: it explains an unfamiliar format, drafts the document you have never written before, and reads a collaborator's configuration back to you in plain language. It is not useful as a source of facts about your data. Every prompt below is assembled with that distinction built in, and each one states the same rules on protected health information, invented identifiers, and the parts a bioinformatician has to confirm.
What the assistant can take, and what stays with a person. Drafting, translating, explaining, and writing the first version of a document or a script are tasks it does well. Deciding what the study asks, approving a mapping or an analysis choice, signing off on anything that reaches a manuscript, and judging what the data mean are not. Paste no protected health information, no participant identifiers, and no small cell counts into any assistant, and treat every output as a draft that a named person reviews.
Your project details
These fill the context block at the top of every prompt. Anything you leave empty is written as a bracketed placeholder so the assistant asks rather than assumes.
Choose a stage of the work
A multimodal project, step by step
Your roadmap
Everything you entered, assembled into one page you can print, download as Markdown, or save to a file and reload later on any machine.