A rare disease program may begin with a known genetic change, a committed patient community, and very little reliable information about how the condition progresses. A clear starting point can coexist with large gaps in the evidence.
AI can help organize scattered observations, compare treatment hypotheses, and reduce repeated computational work. It cannot create missing clinical history or make a small, biased dataset representative by analyzing it more intensively.
The useful question is where these tools can help a team learn more from the research it can realistically do.
The Rare Disease Opportunity
A known disease-associated gene can provide a direction for research. It does not automatically reveal which intervention will help. Researchers still need to understand the gene’s role, the affected tissues, when damage occurs, and whether changing the relevant biology can alter the course of disease.
A molecular effect can also be easier to measure than a clinical benefit. Correcting a laboratory marker is encouraging only to the extent that the evidence connects it with outcomes that matter to patients.
FDA’s rare disease development guidance discusses recurring challenges including limited knowledge, natural history, and clinical development. These questions deserve attention early, alongside candidate discovery.
Understand the Disease Before Comparing Treatments
A natural history study examines how a disease develops over time. It can help identify useful measurements, appropriate observation periods, and differences between patients that a treatment study should account for.
A registry can support that work, but a list of diagnoses is insufficient. The timing and consistency of measurements matter. People seen at specialist centers may differ from those who cannot travel, and people with more severe symptoms may be more likely to have detailed records.
A short observation may capture a substantial change.
A longer follow-up may be needed to observe a difference.
Stability alone would not establish a treatment effect.
Who enters a study and how long they are observed can change the outcome even before any treatment effect is considered.
An external comparison group can sometimes provide useful context, but differences in disease stage, care, measurement, and follow-up can make comparisons misleading. Matching patients on recorded variables cannot guarantee that unrecorded differences have disappeared.
AI can help reconcile terminology or find potentially relevant records. The study still needs a defensible plan for what to compare and how to interpret missing information.
The Platform Advantage
Shared tools can support small programs without making their diseases interchangeable. A team can reuse molecular preparation, property calculations, data checks, and document structures while preserving disease-specific questions.
For example, an analysis workflow may transfer between programs, while its training data or outcome measure does not. Keeping those distinctions explicit prevents a reusable pipeline from becoming an unsupported claim that the scientific problem has already been solved.
At AgentCures, shared computational methods and a version-controlled research record form part of our approach to investigating treatment ideas across programs. The value needs to be assessed through the quality of the evidence produced for each disease.
Drug Repurposing
An existing drug can be a useful starting point when its known activity matches a disease hypothesis. Prior information may help with formulation, exposure, or known adverse effects. The new population, dose, duration, or clinical use can still raise unanswered questions.
Previous formulation work, human exposure, and observed adverse effects in another setting.
What remains to establish?Relevant activity and exposure, suitability for this population, and benefit and risk in the proposed use.
A candidate's activity profile and properties can be designed around the research hypothesis.
What remains to establish?That it can be made, reaches the relevant tissue, has suitable safety properties, and helps in the intended setting.
The SONIA 2 trial illustrates the work involved. Researchers studied nitisinone in alkaptonuria in a four-year randomized trial with 138 participants, comparing treatment with no treatment. They assessed a biochemical endpoint and clinical outcomes, rather than relying on the pathway argument alone.
This was a clinical development example, not evidence of an AI-discovered treatment. It shows why identifying a plausible existing drug and establishing its value for a disease are different research steps.
Patient Impact
Patients and caregivers can identify changes that a convenient laboratory endpoint might miss: fatigue, communication, mobility, pain, or the ability to complete daily tasks. They can also explain which visits and procedures make participation difficult.
That input should influence the research question and measurement plan, rather than appearing only after the protocol has been written. A technically elegant study can be impractical if it requires frequent long-distance travel or measurements that participants cannot complete reliably.
Small datasets also require careful stewardship. Removing a name does not necessarily make a rare combination of clinical details anonymous. Research teams need appropriate permission, access controls, and a plan for what information is shared.
What AI Can Contribute Next
A useful first task is specific and checkable: standardize a set of outcome descriptions, connect a candidate hypothesis to its sources, or identify missing fields needed for a planned analysis. Have someone with disease expertise review the result before expanding the workflow.
Measure useful work completed, including errors corrected. A long list of repurposing suggestions is less informative than a short list with a biological rationale, exposure considerations, contradictory evidence, and a feasible next experiment.
Why We Work on Rare Diseases
The size of a patient population should not determine how seriously its scientific questions are considered. Reusable tools can make some of the supporting work easier for smaller teams.
Progress still depends on understanding the disease and testing treatment ideas carefully. AI contributes when it helps patients, clinicians, and researchers build a clearer account of what is known, what remains uncertain, and which study could resolve the next important question.