Acute Myeloid Leukemia (AML) affects roughly 20,000 people in the US each year and has a five-year survival rate of only about 30%. Despite decades of research and a largely available evidence, identifying which targets are worth pursuing remains difficult. The evidence needed to make that decision is fragmented. General-purpose knowledge graphs are being used to aggregate the information together, but they are not built around the biology and priorities of a specific disease program.

This whitepaper shows how Polly Knowledge Graph connects and grades evidence across biological layers and disease-context, then makes it queryable through AI assistants like Claude, Gemini, ChatGPT via Polly MCP.

In the AML case study, the graph reconstructed the NPM1-mutant differentiation-arrest mechanism and connected it to the approved menin inhibitor revumenib, while also surfacing DOT1L as an investigational opportunity and MEIS1 as an evidence-supported but difficult-to-drug target.

What’s Inside the Whitepaper

• An evidence-centric approach to target discovery: How literature, functional data, disease context, and proprietary evidence are curated and organized into layered knowledge graphs.
• Mechanistic prioritization beyond association scores: How evidence type, tissue relevance, pathways, protein complexes, tractability, and CRISPR dependency are used together to assess targets.
• Natural-language access to the graph through MCP: How scientists can retrieve ranked targets, mechanistic paths, and cited evidence without writing graph queries or knowing the underlying schema.
• A worked AML case study: How six independent evidence layers converged on the MEN1–KMT2A–HOXA9/MEIS1 axis and linked the mechanism to revumenib.
• The next step toward agentic analysis: Temporal disease-state modeling and multi-step agents that can chain biological questions and analyses together.