OnTarget: an agentic system for genetic target triage
Built in partnership with In Your Genes Consulting, OnTarget integrates a growing set of biomedical databases into a single evidence-gathering workflow. Its current operational use is genetics-led target assessment: retrieving relevant records, reconciling evidence across sources, and producing cited outputs for scientific review.
This case study describes the research problem, the system that was built, its effect on the evidence-review process, and the limits of its current scope. A separate technical case study covers the architecture and implementation.
The Client
In Your Genes Consulting
A scientific consultancy led by a PhD geneticist with 15+ years of experience, specializing in early drug discovery, GWAS analysis, and genetic target assessment for biomedical clients.
inyourgenesconsulting.netThe Challenge
Why general-purpose LLMs fall short in genetic evidence review.
Large language models tend to reproduce a familiar human failure mode: a conclusion is formed first and supporting references are located afterward. In biomedical research this ordering is untenable, because the conclusion is only as good as the primary records behind it. The requirement was an evidence-first process in which retrieval precedes synthesis and every assertion remains attached to the record that produced it.
Four practical constraints shaped the design. Relevant genetic data is dispersed across many independently curated databases; assembling it is repetitive manual work; verifying a claim after the fact is labor-intensive; and no consistent procedure existed for applying the same standard across different genetic contexts.
Genetic evidence is only one part of early target assessment. Researchers may also need to understand disease mechanism, tissue and cell-type expression, pathway involvement, tractability, safety liabilities and the existing clinical landscape. The initial OnTarget workflow deliberately addressed a narrower problem: making genetic evidence gathering faster, more consistent and easier to verify.
Information overload
Synthesizing massive, dispersed biomedical datasets during target identification.
Process inefficiency
Bottlenecks caused by manual, repetitive evidence gathering.
Verification gaps
Labor-intensive grounding that leads to inconsistent trust in findings.
Scaling challenges
No consistent, repeatable process across diverse genetic contexts.
The Goals
Three design commitments agreed before implementation.
The scope of the system was fixed at the outset by three requirements, each of which constrained later architectural choices.
Force-multiplier for consultants
Let researchers handle larger, more complex datasets without increasing headcount.
Human-in-the-loop
Automate routine data synthesis so scientists spend their time on interpretation and decisions.
Operational reliability
Perform consistent, repeatable actions — like genetic variant analysis — at scale.
Risks & Scope
Scope and known limitations.
In scope
- Querying biomedical databases (GWAS, eQTL, and more)
- Drafting research reports and automating slide decks
- Synthesizing conclusions from heterogeneous data sources
Out of scope
- —Replacing specialized data science environments such as R-Studio
- —Primary coding work
- —Independent decision-making without human oversight
Known risk
Complex genetic evidence still carries a residual risk of fabricated or misattributed claims. Citation verification is therefore built into the user workflow as an explicit step rather than treated as an optional review.
The Workflow
Target triage and prioritization, step by step.
OnTarget currently supports one workflow end to end: prioritizing candidate targets on the basis of genetic evidence. The sequence below describes how a single question moves through the system.
Frame the question
A consultant states a gene, variant set, or disease hypothesis in natural language. The agent restates the question as an explicit retrieval plan, which the consultant can inspect before execution.
Retrieve genetic evidence
The plan is executed against the integrated databases: association data, variant records and clinical significance, allele frequencies, expression context, regulatory annotation, and supporting literature.
Reconcile and score
Records that describe the same target are aligned across sources, conflicts are surfaced rather than averaged away, and each target receives a suitability assessment with the evidence that produced it attached.
Report with traceable citations
The synthesis is rendered as a brief or slide narrative in which every assertion carries a numbered citation resolving to the primary record, so a reviewer can verify the chain without re-running the search.
Scope of the current system
Genetics-led triage was chosen as the first workflow because it is the client's highest-volume task and because its evidence base is well covered by public databases. Adjacent workflows are under consideration but are not part of the deployed system:
- —Variant and mechanism deep dives, extending the current variant retrieval into structural and functional interpretation
- —Clinical and competitive landscape summaries drawn from trial registries
- —Broader research-brief generation across therapeutic areas beyond genetics-led triage
Development Journey
Four weeks to MVP, twelve to production depth.
The engagement ran in four phases over sixteen weeks, beginning with a working MVP and continuing with iteration on retrieval quality, provenance, and reporting before handover.
Core evidence synthesis
Integrated a growing set of major biomedical databases and trained the agent to select the most effective query strategy for each complex question.
Hardened citations & scale
Iterative releases added bioinformatics visualizations and automated reports, hardened citation mechanisms so every claim stays grounded in human-readable primary sources, and added workflow support for processing hundreds of SNPs at once.
The Outcome
Results after sixteen weeks.
Within its single supported workflow, OnTarget changed the scale at which genetic evidence can be assessed: hundreds of candidate targets can now be evaluated in the time previously required for a handful, with consultant effort concentrated on synthesis and verification.
- Genetic-evidence assessments that previously took weeks are completed in days
- A small consultancy manages larger research programs without adding headcount
- Consultant effort has shifted from manual data-sourcing toward scientific interpretation
- Dispersed genetic datasets are presented as a single synthesized view, reducing time-to-insight
- Hundreds of SNPs are processed within a single workflow run
VERSION 2.0
Lessons learned and future directions.
The first version made clear that the hardest part of the problem is not retrieval but verification: a result is useful only when every claim can be traced back to a primary record. The next version builds on that observation, extending the system from a retrieval workflow toward a collaborative research agent whose reasoning can be inspected, refined, and weighed against expert judgement.
OnTarget assembles and synthesizes evidence relevant to genetics-led target assessment. During review, researchers may encounter questions that fall outside that workflow—for example:
- Whether the implicated gene is active in the relevant tissue or cell population.
- How the target participates in the proposed disease mechanism.
- Whether the target is amenable to an appropriate therapeutic modality.
- Whether existing or discontinued programs reveal efficacy or safety concerns.
OnTarget may retrieve individual records relevant to some of these questions, but they are not currently implemented as validated end-to-end workflows. Their interpretation remains part of the consultant’s broader research process.
Collaborative research agent
Interactive query logs let consultants inspect and refine the agent's reasoning in real time.
Grounded by construction
Claims architected so they are inextricably linked to primary evidence, plus advanced data manipulation primitives.
Cross-model evidence weighting
Evidence weighting aligned with the standards expert human geneticists apply.
Conclusion
A traceable evidence workflow, not autonomous target selection.
OnTarget demonstrates that a substantial part of genetic evidence review can be made repeatable without removing the scientist from the decision process. Its main contribution is not autonomous target selection, but a traceable evidence workflow: questions are decomposed, relevant sources are queried, records remain connected to their provenance, and researchers review the resulting synthesis.
The project also established an important boundary. Integrating a database makes its evidence accessible; it does not make every research question involving that database operationally supported. Each workflow still requires its own scientific framing, evaluation criteria and validation.
OnTarget therefore remains a genetics-led target-assessment system, built on an evidence infrastructure whose components are not inherently limited to that use case.
Could this work for you?
The same pattern in other domains.
Retrieval-first agents with traceable citations are not specific to genetics.