Islet AtlasConnecting early biological changes in type 1 diabetes
A proposed research workspace for following changes over time, understanding what might explain them, and examining the evidence for influencing those processes.
Which changes from a person’s expected biological trajectory reliably precede detectable T1D-related autoimmunity—and which reveal a process we can safely influence?
Design version 1.1 · November 8, 2026 · Open for feedback
Four questions. Different kinds of evidence.
- 01
What changed?
A departure from an expected biological pattern, with its context and uncertainty.
- 02
Does it predict?
Useful information about a defined outcome, tested in independent people.
- 03
What explains it?
Competing mechanisms and evidence that could distinguish them.
- 04
Can changing it help?
Benefit, harms, and durability for a specific intervention and disease stage.
Explore the proposal
Follow a finding through the evidence
Choose a biological signal, then change the disease window or evidence view. Published observations and proposed next studies are labeled separately.
Interactive concept
From early change to a testable mechanism
What changes before detectable autoimmunity?
Follow the timing, the biological explanation, and the evidence for influencing it.
No previously confirmed islet autoantibodies. The first outcome is persistent, confirmed autoantibody detection. Staging definitions (opens in a new tab)
Blood transcriptomics · human longitudinal studies
Type I interferon response
Primary question- Before antibodiesNo confirmed antibodiesSelected window
- Single antibodyNot stage 1
- Stage 1Normal glucose
- Stage 2Abnormal glucose
- Stage 3Clinical T1D
- Observed
- BABYDIET reported a transient, infection-associated RNA signature before first autoantibodies.
- Limiting evidence
- TEDDY found no progression-rate association in 2021; an infection-aligned 2023 analysis found stronger antiviral responses in controls. The signatures and windows differ.
- Window match
- Timing fits this question. Reliable person-level prediction remains to be tested.
Connection to test
Proposed: does unusually slow resolution after an infection distinguish future seroconverters from matched non-converters?
No matched prevention result in this example set
Later-stage treatment findings can suggest experiments. They do not establish benefit before autoantibodies.
A drug affecting a related pathway is not yet a supported treatment connection.
Risk and latest-RNA model versus that same model plus predefined response and recovery features.
First persistent confirmed islet autoantibody; first multiple-autoantibody event analyzed separately.
Trajectory features add no information beyond the latest RNA measurement, or fail to reproduce in an independent cohort.
Linked samples, ages, event intervals, assay metadata, and permission for participant-level analysis.
This concept uses selected published studies. It provides no individual risk estimate. Participant-data analysis and prediction models remain proposed work.
A concrete first test
Does tracking change add useful information?
The proposed first benchmark asks whether a personal molecular trajectory improves prediction of first persistent, confirmed islet autoantibody positivity.
It would compare three models using appropriately accessible longitudinal data. The key comparison is between the latest RNA measurement and that same information plus change over time.
Read the benchmark and validation plan- A
Available risk information
Age, inherited and family risk, and available context.
- B
Add the current molecular state
The same information plus the latest predefined RNA-module values.
- C
Add the trajectory
The same information plus deviations, slopes, persistence, or recovery.
Build this with us
Help shape the next version
Families, people living with T1D, clinicians, researchers, and developers bring different questions. We welcome all of them.
- ClarityWhere does the design need a clearer explanation—or suggest more certainty than the evidence supports?
- Missing evidenceWhich studies, datasets, negative findings, or alternative explanations should be included?
- Research prioritiesWhich proposed question would you most like to see tested, and why?
GitHub feedback is public and requires a GitHub account. Feedback about the design needs no personal health records.