A Knowledge-grounded Agentic Framework for Single-cell RNA Sequencing Method Hypothesis Generation
Year: 2026
Type: preprint
Abstract: Single-cell RNA sequencing (scRNA-seq) analysis relies on computational methods that shape biological conclusions. Developing new methods requires identifying computational operations that address problem constraints to produce intended effects. Here we investigate coordinated agents that extract methodological reasoning from full-text literature and reuse it as prior knowledge for method hypothesis generation. We instantiate this approach as scHypothesizer, a knowledge-grounded agentic framework that uses Mechanism-centered Generation (MCG) to formulate hypotheses from literature-derived Research Primitive Knowledge Graphs (RPKGs) and subjects them to early empirical assessment. We show that coordinated agents recover source-supported reasoning and that full-text RPKGs preserve more complete reasoning than abstract-only representations. In time-filtered retrospective evaluation of 2,429 Problems, MCG achieved the best ranking performance among compared retrieval methods, with a query-macro AUROC of 0.783 and Hit@10 of 69.1%. Finally, scHypothesizer supported early empirical assessment of four researcher-approved hypotheses, providing informative evidence regarding intended Effects, the settings in which they hold, and associated limitations. Together, these results suggest that coordinated agents can make methodological reasoning from prior studies reusable as source-traceable knowledge for generating new method hypotheses and informing their early empirical assessment.
Source: Research Square
Authors: Zhen Chen, Xuewen Zhou, Haiman Guo, Lichao Sun
Institutions: Hong Kong Polytechnic University, Lehigh University
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7236789287
