TriAgent: Automated Biomarker Discovery with Deep Research Grounding for Triage in Acute Care by LLM-Based Multi-Agent Collaboration

TL;DR

This paper tackles the challenge of improving acute-care triage by automating biomarker discovery and grounding findings in biomedical literature. It presents TriAgent, an LLM-based multi-agent framework that combines a data-analysis module (AutoML-driven biomarker identification) with a deep-research module (literature-grounded validation and novelty assessment) to classify biomarkers as grounded or novel, with auditable justifications. Empirical results show TriAgent outperforming single-agent and prompting baselines on topic adherence and faithfulness across foundation models, with an ablation indicating an optimal sub-agent count around five and clear trade-offs between speed and cost. The work advances transparent, end-to-end biomarker discovery for real-time acute-care triage, and points to future clinical validation, broader multi-center data, and improved evidence retrieval for practical deployment.

Abstract

Emergency departments worldwide face rising patient volumes, workforce shortages, and variability in triage decisions that threaten the delivery of timely and accurate care. Current triage methods rely primarily on vital signs, routine laboratory values, and clinicians' judgment, which, while effective, often miss emerging biological signals that could improve risk prediction for infection typing or antibiotic administration in acute conditions. To address this challenge, we introduce TriAgent, a large language model (LLM)-based multi-agent framework that couples automated biomarker discovery with deep research for literature-grounded validation and novelty assessment. TriAgent employs a supervisor research agent to generate research topics and delegate targeted queries to specialized sub-agents for evidence retrieval from various data sources. Findings are synthesized to classify biomarkers as either grounded in existing knowledge or flagged as novel candidates, offering transparent justification and highlighting unexplored pathways in acute care risk stratification. Unlike prior frameworks limited to existing routine clinical biomarkers, TriAgent aims to deliver an end-to-end framework from data analysis to literature grounding to improve transparency, explainability and expand the frontier of potentially actionable clinical biomarkers. Given a user's clinical query and quantitative triage data, TriAgent achieved a topic adherence F1 score of 55.7 +/- 5.0%, surpassing the CoT-ReAct agent by over 10%, and a faithfulness score of 0.42 +/- 0.39, exceeding all baselines by more than 50%. Across experiments, TriAgent consistently outperformed state-of-the-art LLM-based agentic frameworks in biomarker justification and literature-grounded novelty assessment. We share our repo: https://github.com/CellFace/TriAgent.

Figures (5)

• Figure 1: Architecture of TriAgent. TriAgent is a graph-based multi-agent framework for automated biomarker discovery and literature-grounded validation. The Scoping and Data Analysis Agents produce structured briefs (Brief-1, Brief-2); the Research Supervisor Agent coordinates sub-agents for RAG-based searches; the Reporting Agent integrates findings into an auditable report.
• Figure 2: Evaluation of the effect of varying the number of research sub-agents on grounding quality and novelty assessment with the performance.
• Figure 3: Heatmap highlighting the feature importances across multiple trained models by the Data Analysis Agent.
• Figure 4: Identified features and their respective relative importances obtained from the best performing model, Gradient Boost Machine (GBM), achieving a root mean square error of 0.2065.
• Figure 5: The SHAP explanations showing feature importance and normalized feature-value distributions for infection prediction by the GBM model.