Receptor signaling
Think of a receptor receiving a signal. GLP‑1 receptor agonists work by activating that receptor.
Example Specific GLP‑1 medicines have U.S. approval for specific indications.
Read the GLP‑1 guideFREE ACCESS / LEARNING GUIDE
Turn a question about your body into a tool you can use.
Start with GLP‑1, RNA, and genome editing. Check a piece of evidence, use Codex or Claude Code to build a tool, then test the result yourself.
Browse without signing up. Workshops use synthetic data and your own AI agent.
01 START HERE
Three layers of technology
Start by asking where an intervention acts and what evidence supports it. Check the study stage, indication, and jurisdiction separately.
RNA interventions
Some RNA interventions affect how cells use a working copy of genetic instructions. RNA technologies act through several mechanisms.
Example Inclisiran, a small interfering RNA medicine, has U.S. approval to lower LDL cholesterol in specific populations.
Read the FDA approval exampleDNA editing
Changing source code is a metaphor for altering a DNA sequence. Gene therapy also includes approaches that introduce genetic material.
Example Casgevy has U.S. approval for certain blood disorders. The product information defines who is eligible.
Read the FDA product informationThis guide uses “body coding” as a teaching metaphor. Sequencing reads data. A simulation calculates outcomes under defined conditions.Try a simulation
MODEL ≠ MEASUREMENT
How can you check AI outputs?
- Structure prediction
- AlphaFold predicts protein structures from sequences. Compare predictions with structures measured in experiments. Clinical effects are evaluated in separate studies. Read the EMBL-EBI explainer
- Virtual cells
- Research models estimate cell responses under specified conditions. To evaluate the predictions, use experimental data the model has never seen. Check which cell types and conditions the findings cover. Read about the Arc initiative
- Coding agents
- Codex or Claude Code can turn your tracking requirements into a webpage. Start with examples that have known answers, then check the sorting, missing values, and print output. Try the checks in Lesson 1
02 BUILD WITH AGENTS
Build with your AI agent
Start with a question, open a finished example, and choose one thing to change. Use the sample data to check the result.
1 Choose a question to investigate
2 Have your agent build the tool
3 Check it against the sample data
01Build a GLP‑1 follow-up logStart with dates, symptom notes, weight records, and questions for a clinician.Beginner
Result An offline timeline and printable summary. Open the demo and decide what to add. For example, you can prioritize symptom notes on screen and keep all records in print.
- Create an empty folder and put the downloaded CSV in it.
- Open the folder with your AI agent. Get started with Codex · Get started with Claude Code
- Pick a customization and copy the updated prompt. You can also edit the last line yourself.
- Double-click the generated
index.html, then check it against the list below.
Put only the fictional CSV in the project folder. Keep real medical records, test reports, DNA files, and identifying information outside the exercise.
Send to your agent
Build a follow-up log in the current folder that I can open by double-clicking. Read sample-followup.csv and embed the eight fictional rows in the finished page so it runs offline. If the file is missing, ask me to place the CSV in the folder.
Use plain HTML, CSS, and JavaScript. Show the date, weight in kilograms (which may be blank), symptom note, and follow-up question in ascending date order. Plot weight and break the line where a value is missing. Include all eight records, original units, and questions in the printout. The page organizes sample records and gives no diagnosis, medication, or dosing advice.
Check that September 10 comes before September 17, the missing September 24 weight stays blank, and the printout contains eight records. Tell me how to open index.html.
Customization
Add a notes-only view and keep all eight records in print.- September 10 comes before September 17.
- The September 24 weight stays blank, and the chart line breaks there.
- The printout has all eight records.
If no data appears, check the CSV filename and the folder where your agent saved the output. You can explore the finished demo while setting up your agent.
02Build a genetics simulation with synthetic dataUse fictional groups to see why associations can change after splitting a dataset.Intermediate
Result A classroom demo with two views. Use the same 80 fictional records to check what you conclude from pooled and grouped results.
Make a guess and compare the groupsSend to your agent
Build an offline learning page with plain HTML, CSS, and JavaScript. Generate 80 fixed fictional records with a batch (P or Q), an abstract marker (X or Y), and a fictional score. Batch P has 30 X and 10 Y records, all with a score of 70. Batch Q has 10 X and 30 Y records, all with a score of 30. Ask readers for a prediction, then let them switch between Overall and By batch. Calculate counts and averages from the same array. Explain how batch composition changes the pooled average. Use abstract markers and keep real DNA, personal risk estimates, and intervention advice out of the exercise. Verify that X averages 60 and Y averages 40 overall, both average 70 in batch P, and both average 30 in batch Q.
Check There are 80 records. The overall marker averages differ by 20 points, and the marker averages are equal within each batch.
03Build a research evidence bookmarkConnect news, trial registrations, and original announcements.Beginner
Result A personal research bookmarks page. Practice with Fractyl RJVA‑001 and use a clinical trial registry and a company disclosure to answer two questions. What did the registry record? What decision did the company make?
Send to your agent
Build a static research bookmarks page. Each project can have multiple evidence entries. Include the date, jurisdiction, evidence type, original link, what is known, and what needs verification. Start with three placeholder entries for RJVA-001 covering trial registration, the company decision, and findings from human studies. I will add verified sources manually. Allow HTTP and HTTPS links only. Show Unverified for entries without a source. Add a filter by evidence type. Check that several sources can appear for one project, an empty filter gives a clear message, and a human result without a source remains unverified.
Check The two sources keep their own dates and evidence types. Results without a source remain marked Unverified.
03 FIELD NOTES
Research updates
Each update gives its evidence type, scope, and original source.
Sources checked
CTX310 early human results
An early trial observed reductions in blood lipid measures in participants with severe hypertriglyceridemia. The company disclosed findings from a small group. Long-term clinical benefit still needs study.
Phase 1b · Company announcementOriginal source
FDA requires updated GLP‑1 drug labeling
The U.S. FDA required labels for affected medicines to include a potential risk of NAION, a rare optic nerve disorder. The original communication describes the products and evidence it covers.
United States · Drug safety communicationRead the evidence card
Fractyl stops internal development of Rejuva
Fractyl disclosed that it would stop internal development and explore alternatives. The platform's RJVA‑001 program targets type 2 diabetes. Track the company decision, trial registration, and human findings as separate records.
Company disclosure · Development updateOriginal source
Lecheng project reports treatment of its first patient
Divine Hospital reported local progress on a GLP‑1/FGF21-related cell therapy project. The announcement records a treatment milestone. Assessing efficacy requires separate evidence.
China · Hospital announcementOriginal source
Reading a chart · How do drug sales compare with annualized AI revenue?
The video puts GLP‑1 sales and annualized AI company revenue on the same chart. Check the time period, the products or companies included, and whether the numbers represent actual sales, analyst estimates, or annualized revenue based on a recent period. Use those answers to decide whether a ratio between the bars would be meaningful.
See the chart in the video · Check what the market figures cover