"My patient finally has a genetic diagnosis... but how do I treat it?"
~ Clinician caring for a rare disease patient
Too often, a genetic diagnosis is just the beginning of a long, uncertain search for therapies. The SAIL AI-Tx Challenge calls on innovators to build AI systems that turn diagnoses into actionable treatment options for rare disease patients.
Final Challenge · Now Open Actionability Report Generation
Turn a rare disease diagnosis into a comprehensive, patient-specific therapeutic actionability report — established therapies, trial eligibility, off-label options, and bespoke strategies, all with traceable evidence.
Submit your model → Deadline: November 1, 2026
The Challenge: Actionability Report Generation
The Actionability Report Generation task is now open to submissions from any participants! This final task tests an AI system’s ability to generate a comprehensive therapeutic actionability report for a hypothetical rare disease patient. This task will require open-ended narrative synthesis covering established and supportive therapies, clinical trial eligibility, emerging bespoke approaches (e.g., ASOs, gene therapies), and repurposing candidates. Details about the expected input format, required output format, and evaluation details are below.
This challenge was previously planned as an invite-only task for top performers from an earlier challenge phase, but any team may now participate. Hypothetical patients will be released on October 1, 2026.
AI models will be provided with an input JSON object containing entries for hypothetical patients, including 1+ genetic variants and brief clinical context. Input tasks will be formatted as follows:
| Field | Description |
id (string) | Unique identifier for the task instance (used to link input and output). |
patient.genotype (array of objects) | List of one or more variant objects, where each object includes: -
gene (string) – HGNC gene symbol -
transcript (string) – RefSeq or Ensembl transcript identifier -
variant_cdna (string) – cDNA HGVS notation -
variant_protein (string) – Protein HGVS notation -
zygosity (string) – e.g., "heterozygous", "homozygous", or "hemizygous" Users can assume that patient.genotype corresponds to a true diagnosis. |
patient.clinical_context (string) | Brief free-text clinical description. |
For example,
{
"id": "AITX-30001",
"patient": {
"genotype": [
{
"gene": "CFTR",
"transcript": "NM_000492.4",
"variant_cdna": "c.1521_1523del",
"variant_protein": "p.Phe508del",
"zygosity": "homozygous"
}
],
"clinical_context": "A 15-year-old male with a diagnosis of cystic fibrosis presents with chronic sinopulmonary disease characterized by persistent respiratory symptoms and recurrent infections. He also has exocrine pancreatic insufficiency requiring enzyme replacement therapy and exhibits features of male infertility consistent with congenital bilateral absence of the vas deferens (CBAVD)."
}
}
Submitted AI models must return a structured, markdown-formatted actionability report as a string within a JSON object. Details below.
| Field | Description |
id (string) | Must match the corresponding input task id. |
report_markdown (string) | Therapeutic actionability report following the required template. Subheadings, bullets, links, and references beyond the template are encouraged. |
Markdown Report Template
These requirements apply to every section of the report:
1 Cite a verifiable source for every claim, or show your reasoning. A source need not be a paper! A database record, drug label, guideline, or trial registry entry is fine, given a URL or accession. If no prior report(s) exist, cite the underlying facts and lay out the reasoning that gets you to the claim. Never cite a reference that doesn't support the claim.
2 State conditional actions. If missing patient information would change a recommendation, say so: "If seizures occur more often than once every three days, then XXX. Otherwise, YYY."
3 Don't repeat across sections. Put each action in the single best-fitting section. An ASO approved for the patient's specific variant and presentation belongs under Targeted Therapies, not again under Bespoke Therapies.
Every report must contain the following sections, in order:
Opens the reportSummary
The highest-value actions for this patient, ordered by tier, including conditional actions depending on patient-specific clinical variables. If key information that would alter therapeutic options is missing, this should be explicitly noted.
Tier 1Established Actions
Targeted Therapies Approved or guideline-supported treatments directed at the underlying molecular defect or disease mechanism.
Supportive Management Standard-of-care pharmacologic or procedural management of symptoms and complications rather than the underlying defect. Surveillance is currently excluded.
Contraindications & Precautions Agents, doses, or procedures to avoid or modify because of the variant or disease, including anesthesia risks and clinically relevant medication interactions.
Tier 2Emerging Actions
Clinical Trial Eligibility Actively recruiting clinical trials for which the patient may qualify, with potential eligibility assessed against the available clinical and molecular criteria.
Off-label or Expanded-Access Therapies Existing therapies that could potentially be accessed outside standard approval for this indication, and for which human treatment precedent exists, such as at least one published report of the therapy administered to a patient with the same gene, disease, or variant context.
Tier 3Hypothesized Actions
Bespoke Therapies Assessment of the variant's amenability to specific DNA- and RNA-directed therapeutic strategies (e.g., ASOs, gene therapies) not already identified as established or emerging, with a verdict and rationale for each.
Drug Repurposing Existing approved or investigational compounds for which no such human treatment precedent exists, but with a mechanistic rationale linking the intervention to the molecular defect.
For example,
{
"id": "AITX-30001",
"report_markdown": "## Therapeutic Actionability Report for Patient AITX-30001\n\n### Summary..."
}
3 Evaluation
Particularly challenging patient cases and anticipated answer keys are being developed by a panel of experts in clinical genetics, translational research, and drug development. User model submissions will be evaluated by these same expert panels. Reports will be scored based on the following dimensions:
Expert panel reviewScoring Criteria
Factual Accuracy Scientific and clinical accuracy of all claims in the response. Incorrect and misleading interpretations will be scored negatively, even if all other required key facts are present and correct.
Completeness Coverage of all prespecified report sections and all expected elements within each section, including whether relevant therapeutic actions, precautions, and uncertainties are identified and adequately addressed.
Evidence & Justification Whether conclusions are logically supported by appropriate, traceable evidence and/or key sources and citations, and whether those claims can actually be verified.
Prioritization & Calibration Whether actions are appropriately prioritized and placed within the prespecified evidence tiers (Established, Emerging, Hypothesized). Confidence and speculation should reflect the strength of the supporting evidence, with major uncertainties explicitly acknowledged.
Clinical Utility Whether findings are clearly translated into patient-specific actions or next steps.
Tutorials: links to databases and video tutorials on utilizing compute resources.
FAQ: responses to questions regarding challenge scope and requirements.
