Activities
2026
POSTER
Locally Deployed Agentic LLMS for Structured Data Extraction From Complex Oncology EHRS
Event:
AI for Oncology and Cancer Research. May 7th - 8th, 2026. Milan, Italy
Abstract:
Background: Large language models (LLMs) have potential for structuring clinical data, yet extracting structured variables from lengthy EHRs, especially in non-English settings, remains challenging. We evaluated the reliability of open-source LLMs for this task using complex oncological EHRs while ensuring data privacy through on-premise deployment.
Methods: We studied a dataset from St. Olavs University Hospital, including 62 patients treated for stage IV colorectal cancer from 2010 to 2020, with EHRs ranging from 180,000 to 1.7 million characters. We manually extracted fifty variables related to patient characteristics and treatments to establish a gold standard, benchmarking extraction strategies with models under 100B parameters.
Results: The sequential approach achieved an F1 score of 0.74 and an accuracy of 0.73, with categorical variables like BRAF mutation status (F1 = 0.983) performing best, while temporal and numerical variables were more challenging. We developed a retrieval framework using MCP search, resulting in roughly 50-fold faster processing and improved accuracy for temporal variables.
Conclusions: Locally deployed LLMs can effectively extract structured oncology variables from complex, non-English EHRs. Improved retrieval architectures demonstrate potential for enhancing speed and performance, with ongoing work aimed at optimizing methods and expanding the dataset.
Presenter:
Andreas Eikså
Thomas Falkeid Hagland
Leo Quentin Thorbjørnsen Bækholt
2025
POSTER
Model-Guided Discovery of Synergistic Drug Combinations in Colorectal Cancer
Event:
Computational Systems Biology of Cancer 8th edition. Sept. 22nd - 26th, 2025. Institut Curie, Paris
Abstract:
Nearly half of colorectal cancer (CRC) patients have druggable molecular alterations, but single-target therapies often fail due to resistance. To improve treatment strategies, we used a Boolean modeling approach to identify effective drug combinations for CRC cell lines. From 880 candidate combinations, we validated 176, with 119 confirmed as synergistic, achieving a 68% success rate—significantly higher than past efforts.
Our framework demonstrated strong predictive performance with 72% accuracy, 76% precision, and 87% recall, correctly identifying almost 90% of observed synergies. Most successful combinations involved mTOR or AKT inhibitors, while combinations with CHEK1 inhibitors were less effective. We discovered two novel synergies: mTOR with ERK and mTOR with BCL2, both confirmed across 16 cell lines and supported by analysis suggesting mechanisms of action.
These findings highlight the potential of our modeling approach to enhance drug discovery efficiency, prioritize effective combinations, and deepen our understanding of CRC biology.
Presenter:
Viviam Bermúdez Paiva
2024
POSTER
From data integration to actionable insights: Optimization of model-guided drug synergy predictions in systems medicine
Event:
Norwegian Bioinformatics Days 2024. May 28th - 30th, 2024. Bergen, Norway.
Abstract:
Functional precision medicine presents a significant opportunity in oncology by enabling the design and testing of treatments using patient companion models. A key challenge is creating efficient computer models to select optimal therapies from limited patient-derived material. This project seeks to develop a platform for customizing disease Boolean models into cell-line, organoid, and patient-specific models to simulate candidate drug effects. We utilized bioinformatics and machine learning to integrate diverse datasets for model customization and drug panel structuring. Synergy predictions were computed using the DrugLogics pipeline for each calibrated model, with predictive accuracy evaluated against experimental data. The integration of these datasets facilitates a comprehensive analysis of model performance and synergy predictions, leading to biological hypotheses about drug synergies. Our results showed that, for a publicly available dataset on colon cancer cell lines, our models achieved an average ROC AUC of 0.59 and a highest value of 0.78 for cell line SW48, indicating successful differentiation of true synergistic combinations. In summary, this work enhances in vitro testing and model-driven decision-making for personalized treatment plans, expanding treatment options and improving insights into disease biology while leveraging the NTNU DrugLogics pipeline.
Presenter:
Viviam Bermúdez Paiva
2023
POSTER
The bumpy road towards a digital-twin drug-prediction platform
Event:
PerMedCoE summer school: from pathway modelling tools to cell-level simulations. June 25th - 30th, 2023, Barcelona, Spain.
Abstract:
Functional precision medicine offers a crucial opportunity in oncology, allowing treatment design and testing in patient companion models. However, the key challenge lies in developing efficient computer models to select optimal candidate therapies that can be tested in limited patient-derived material. Our project aims to develop a platform that enables the customization of Boolean models into patient-specific models to simulate the action of candidate drugs. We employ bioinformatics, artificial intelligence, and machine learning approaches to integrate and harmonize datasets on available drugs, drug targets, existing drug synergy data, and precise activity states of key regulatory network proteins derived from patient omics data. The model’s predictive accuracy for drug responses enables in vitro testing and data-driven decisions for tailored treatment plans, considering a wide range of treatment designs and approved drugs, including repurposed and novel combinations. Through the fusion of diverse biological data, model simulations, and prediction analysis, the platform serves as a tool within the NTNU DrugLogics pipeline, paving the way in the field of precision oncology.
Presenter:
Viviam Bermúdez Paiva
