Process-Supervised Multi-Agent Reinforcement Learning for Reliable Clinical Reasoning

Clinical decision-making requires nuanced reasoning over heterogeneous evidence and traceable justifications. While recent LLM multi-agent systems (MAS) show promise, they largely optimise for outcome accuracy while overlooking process-grounded reasoning aligned with clinical standards. One critical real-world case of this is gene-disease validity curation, where experts must determine whether a gene is causally implicated in a disease by synthesising diverse biomedical evidence. We introduce an agent-as-tool reinforcement learning framework for this task with two objectives: (i) process-level supervision to ensure reasoning follows valid clinical pathways, and (ii) efficient coordination via a hierarchical multi-agent system. Our evaluation on the ClinGen dataset shows that with outcome-only rewards, MAS with a GRPO-trained Qwen3-4B supervisor agent substantially improves final outcome accuracy from 0.195 with a base model supervisor to 0.732, but results in poor process alignment (0.392 F1). Conversely, with process + outcome rewards, MAS with GRPO-trained supervisor achieves higher outcome accuracy (0.750) while significantly improving process fidelity to 0.520 F1. Our code is available at https://github.com/chaeeunlee-io/GeneDiseaseCurationAgents.

GNN-Suite: a Graph Neural Network Benchmarking Framework for Biomedical Informatics

We present GNN-Suite, a robust modular framework for constructing and benchmarking Graph Neural Network (GNN) architectures in computational biology. GNN-Suite standardises experimentation and reproducibility using the Nextflow workflow to evaluate GNN performance. We demonstrate its utility in identifying cancer-driver genes by constructing molecular networks from protein-protein interaction (PPI) data from STRING and BioGRID and annotating nodes with features from the PCAWG, PID, and COSMIC-CGC repositories. Our design enables fair comparisons among diverse GNN architectures including GAT, GAT3H, GCN, GCN2, GIN, GTN, HGCN, PHGCN, and GraphSAGE and a baseline Logistic Regression (LR) model. All GNNs were configured as standardised two-layer models and trained with uniform hyperparameters (dropout = 0.2; Adam optimiser with learning rate = 0.01; and an adjusted binary cross-entropy loss to address class imbalance) over an 80/20 train-test split for 300 epochs. Each model was evaluated over 10 independent runs with different random seeds to yield statistically robust performance metrics, with balanced accuracy (BACC) as the primary measure. Notably, GCN2 achieved the highest BACC (0.807 +/- 0.035) on a STRING-based network, although all GNN types outperformed the LR baseline, highlighting the advantage of network-based learning over feature-only approaches. Our results show that a common framework for implementing and evaluating GNN architectures aids in identifying not only the best model but also the most effective means of incorporating complementary data. By making GNN-Suite publicly available, we aim to foster reproducible research and promote improved benchmarking standards in computational biology. Future work will explore additional omics datasets and further refine network architectures to enhance predictive accuracy and interpretability in biomedical applications.

The Ontoverse: Democratising Access to Knowledge Graph-based Data Through a Cartographic Interface

As the number of scientific publications and preprints is growing exponentially, several attempts have been made to navigate this complex and increasingly detailed landscape. These have almost exclusively taken unsupervised approaches that fail to incorporate domain knowledge and lack the structural organisation required for intuitive interactive human exploration and discovery. Especially in highly interdisciplinary fields, a deep understanding of the connectedness of research works across topics is essential for generating insights. We have developed a unique approach to data navigation that leans on geographical visualisation and uses hierarchically structured domain knowledge to enable end-users to explore knowledge spaces grounded in their desired domains of interest. This can take advantage of existing ontologies, proprietary intelligence schemata, or be directly derived from the underlying data through hierarchical topic modelling. Our approach uses natural language processing techniques to extract named entities from the underlying data and normalise them against relevant domain references and navigational structures. The knowledge is integrated by first calculating similarities between entities based on their shared extracted feature space and then by alignment to the navigational structures. The result is a knowledge graph that allows for full text and semantic graph query and structured topic driven navigation. This allows end-users to identify entities relevant to their needs and access extensive graph analytics. The user interface facilitates graphical interaction with the underlying knowledge graph and mimics a cartographic map to maximise ease of use and widen adoption. We demonstrate an exemplar project using our generalisable and scalable infrastructure for an academic biomedical literature corpus that is grounded against hundreds of different named domain entities.

Edinburgh Clinical NLP at MEDIQA-CORR 2024: Guiding Large Language Models with Hints

The MEDIQA-CORR 2024 shared task aims to assess the ability of Large Language Models (LLMs) to identify and correct medical errors in clinical notes. In this study, we evaluate the capability of general LLMs, specifically GPT-3.5 and GPT-4, to identify and correct medical errors with multiple prompting strategies. Recognising the limitation of LLMs in generating accurate corrections only via prompting strategies, we propose incorporating error-span predictions from a smaller, fine-tuned model in two ways: 1) by presenting it as a hint in the prompt and 2) by framing it as multiple-choice questions from which the LLM can choose the best correction. We found that our proposed prompting strategies significantly improve the LLM's ability to generate corrections. Our best-performing solution with 8-shot + CoT + hints ranked sixth in the shared task leaderboard. Additionally, our comprehensive analyses show the impact of the location of the error sentence, the prompted role, and the position of the multiple-choice option on the accuracy of the LLM. This prompts further questions about the readiness of LLM to be implemented in real-world clinical settings.