Models that showtheir reasons.
The ranking models suggest where to look and give reasons you can check against the sources. None of them gives a diagnosis, and none gives the chance that a gene is the answer.
Phenotype ranker
Compares the patient’s phenotype with what is known about genes and diseases, and ranks the genes whose profile fits best. A classical HPO similarity score runs next to the learned ranker as an independent check. Absent, unknown and not tested are kept apart.
Genotype ranker
Weighs the patient’s rare variants together with the phenotype score: how common they are, their predicted effect and how they are inherited in the family. The reasons are listed with each candidate.
New-gene ranker
For cases where no established gene fits. It ranks genes with little or no established disease link by how well everything else known about them fits the patient. They appear as a separate list, never mixed in with established findings, and each comes with its reasons.
Note reader
Reads referral letters and clinic notes and suggests the phenotype terms they describe, each with the sentence it came from. Hedged or ambiguous statements, such as “possible seizures” or a symptom that may belong to a relative, are flagged for your decision with the reason.
Investigation agents
A set of specialised agents works through a case: comparing hypotheses, collecting evidence, searching the literature and re-checking cases when knowledge changes. Each agent has a fixed remit enforced by the servers, not by its prompt. They can propose; they cannot approve or sign. Every step is recorded.
Case assistant
Ask in plain language, such as “why is this gene ranked third?” or “what would rule it out?”. Answers draw on the case’s own record and cite what they used. A statement that cannot be checked is marked as not verified.
The whole genetic picture.
Works from VCFs and sequencing files, for single patients and families.