.webp)
.webp)
.webp)
.webp)
.webp)
Every connection in Polly KG is a specific, labeled claim, not a guess based on what showed up in the same paper.
Once the graph is built on direction-graded evidence, your target selection runs on the same standard.
Each relationship carries a type, a direction, and a polarity score before it becomes an edge. Contradicting evidence surfaces during scoring, before it costs a validation cycle.
.svg)
Custom scoring across disease relevance, evolutionary conservation, and druggability narrows a full gene set to a ranked, defensible list.
Scout-sourced evidence and Xtract-structured proprietary data enter the same graph, so spatial, GWAS, and patient-stratification work start from one consistent schema.
.webp)
.webp)
.svg){kind=small}
Polly KG is the evidence backbone inside both of Elucidata's partnership models.
.svg)
Elucidata leads target discovery and uses Polly KG for the customer, scoring each relationship from nomination to validation so contradictions appear before committing to a candidate. Customers get a concise, evidence-backed shortlist.
.svg)
When Elucidata owns a customer's computational infrastructure layer, whether that's a knowledge graph deployment, an omics pipeline environment, or an AI co-scientist workflow, Polly KG usually sits underneath it, direction-graded from the day it's built rather than qualified after the fact.
We integrate Polly KG into your target ID infrastructure in the model that fits your team.
Web interface for KG exploration, target scoring dashboards, and evidence report generation via Polly Lens.
Connects your existing AI tooling directly to Polly KG. List graphs, run Cypher queries, download results.
Query Polly KG directly from Python, or wire it into your stack through the Polly KG REST API, with Cypher and natural-language queries supported and async execution for large result sets.
A guided AI research interface for querying the graph, exploring mechanisms, and generating evidence reports without writing a query. Also available through Slack.
Full API documentation and integration guides available on request.
Co-occurrence counts how often a gene and a disease appear together in the same text, with no direction, no polarity, and no mechanism. Polly KG grades every relationship as supporting, contradicting, or neutral, with a direction and a certainty level, before writing it as an edge. For target identification, that distinction determines which candidates make the shortlist.
Target ranking narrows a full gene set, commonly around 20,000 candidates, to a scored shortlist using the graph. Target validation uses the same qualified evidence to stress-test specific candidates before a wet-lab commitment. Both run on direction-graded evidence. The question changes from "who should we pursue" to "does this candidate hold up."
It gives your scoring layer direction-graded evidence for each candidate. Candidates with mixed or contradicting evidence are flagged before they reach the shortlist.
Yes. Full-text processing recovers evidence from methods sections and supplementary data that abstract-only tools miss, roughly 8 million full-text papers Open Targets' abstract-based evidence layer doesn't reach. Proprietary in-house datasets go through the same qualification pipeline as public literature. That pipeline let one program surface a viable target in a disease area with only a single public dataset available.
Literature is ingested continuously. Proprietary data integrates at your program's milestones.
The full range, from initial target discovery through target validation. The same graph that ranks a full gene set also carries the contradicting evidence your team needs at the validation stage.
AI models that query a co-occurrence graph amplify its noise. Polly KG gives those models qualified, direction-graded inputs, so predictions start from defensible evidence.
.svg)
.svg)
.svg)
%201%20(1).svg)
%20(1).svg)
%201%20(1).svg)
.svg)
%201.svg){kind=icon}
.webp)
.webp)
.svg)
.svg)
