5P - Onconaut: A precision medicine platform for oncology therapies

Background

The advent of precision medicine has significantly altered the landscape of cancer treatment, introducing a new era where therapies are customized to the individual's molecular profile. In this context, we introduce Onconaut, an AI-driven platform engineered to facilitate the selection of targeted oncology therapies for patients and caregivers.

Methods

Onconaut leverages advanced language models and machine learning techniques to evaluate patient data, aligning patients' molecular profiles with the most suitable therapies and clinical trials. The platform integrates 30M+ records from PubMed database, thousands of clinical trials, curated biomarker-treatment sets and oncology guidelines. Given a patient profile, onconaut uses multiple language models to integrate date from the sources mentioned above.

Results

The implementation of Onconaut has shown promise in simplifying the treatment selection process, enabling the identification of personalized therapy options that are closely aligned with the unique genetic makeup of a patient's cancer. This method has the potential to enhance treatment efficacy while reducing the likelihood of adverse reactions and ineffective treatments. Additionally, Onconaut serves as a conduit for patients and clinicians to stay abreast of the latest developments in biomarker research and its implications for treatment strategies. We have benchmarked Onconaut's clinical trial matching ability against available tools and we have seen 2x improvement in accuracy. In addition, we have also benchmarked guideline-based treatment selection strategy against real-life and synthetic datasets. Again, improvement of onconaut against off-the-shelf AI-tools is clearly demonstrated.

Conclusions

Onconaut stands at the forefront of precision medicine, offering a transformative tool for navigating the complexities of targeted cancer therapy. By utilizing language models and machine learning technologies, it offers tailored treatment recommendations, finds relevant clinical trials, and delivers information on biomarkers. Although, we were able to show superiority over other tools, the platform's accuracy is constanly tested against newly available methods and improved.

Editorial acknowledgement

Clinical trial identification

Legal entity responsible for the study

The author.

Funding

Helmholtz Association.

Disclosure

The author has declared no conflicts of interest.