Omics data to target prioritization
Use a transcriptomic dataset and its experimental context to produce five ranked target hypotheses, each carrying supporting and contradicting evidence, an explicit uncertainty statement, and a recommended validation experiment.
GSE54456
10
5
21
RNA-seq from lesional skin biopsies of 92 psoriasis patients and normal skin from 82 individuals. The associated study reported 3,577 differentially expressed genes (1,049 up, 2,528 down) and used pathway and weighted co-expression analyses to investigate immune and epidermal programs.
The published study aimed to understand disease mechanisms. It did not perform target nomination. The five-target prioritization objective used here is an illustrative extension, not a result of the published work.
fig. 03 — the run, stage by stage
Walk the ten stages the way the system would
Every stage carries its inputs, its methods and the capability class each one belongs to, what it writes to scientific memory, and what it contributes to the work product. Watch for the places the run could have gone wrong: quality control naming a tissue-composition confounder before anything is interpreted, evaluation separating whether the signal is robust from whether it supports a mechanism, and a rubric whose weights change with the objective.
stage 01 · Interpretation
Define the biological and therapeutic objective
What are we trying to decide, and what would a good answer look like?
architecture layer
Scientific objective
capability classes
- Disease area and therapeutic hypothesis
- Decision the work feeds and its owner
- Modality and platform constraints
- Required output format and review standard
Objective structuring
Loop
Therapeutic strategy context
Org
- Structured scientific objective
- Required evidence dimensions
- Named reviewer and decision owner
- Disqualifying conditions
- Objective statement
- Success criteria and disqualifying conditions
- Decision owner and reviewer
- Genetics-anchored objective: prioritize causal human genetic support
- Modality-anchored objective: prioritize extracellular accessibility for an antibody program
- Confirms the objective, the required evidence dimensions and the standard the memo must meet before any compute is spent.
- 01Is the objective specific enough to constrain method selection?
- 02Are success criteria stated before any analysis runs?
- 03Is the decision the work feeds named, with an accountable owner?
The memo opens with the objective it was written to answer, so a reviewer can judge whether it answered that question.
- 01Objective statement
- 01Success criteria and disqualifying conditions
- 01Decision owner and reviewer
These are not notes. They are the premises that constrain what the AI scientist may conclude at later stages, and compression must preserve every one.
- 01The memo opens with the objective it was written to answer, so a reviewer can judge whether it answered that question.
Assembled throughout the loop. That is what makes claim-level provenance possible.
How a general AI scientist emerges from one loop
Every method above is one of four kinds. The first two are reusable assets that compound across loops. The second two are what make this loop yours.
Generic AI scientist capability
Reusable infrastructure. Built once, used by every work loop — runtime, memory, provenance, orchestration.
Scientific skill
A validated scientific method, reusable across related scientific problems. Versioned and testable, like a library function for science.
Loop-specific configuration
Unique to one work loop. The recipe, thresholds and rubric that make omics-to-target different from safety-signal investigation.
Organization-specific context
Your therapeutic strategy, internal datasets, platform capabilities, prior failures and scientific preferences. Cannot be bought off the shelf.
How mature is the system you use?
Assess a product, an internal system or your organization against the capability model this loop depends on.