Each year, nearly fifty new cancer therapies are approved worldwide. This offers new hope for patients, but it is becoming increasingly difficult for doctors to make the right choice from the growing range of treatments available. After all, every tumour has unique biological characteristics, which means that finding the most effective therapy requires a tailored approach. To support this process, a research team at the Max Delbrück Centre (MDC-BIMSB) in Berlin has developed a new AI toolkit: Flexynesis.
Flexynesis uses both classic machine learning and advanced deep learning methods. This enables the tool to analyse different types of data simultaneously: from genetic and molecular data (multi-omics) to medical images and clinical reports. According to project leader Dr Altuna Akalin, this enables doctors to make more accurate diagnoses and prognoses, and to choose more targeted treatment strategies.
Flexibility and broad applicability
The possibilities of Flexynesis are broad. For example, the tool can help determine the type of cancer, predict the effectiveness of specific drugs and estimate a patient's chances of survival. In addition, the toolkit supports the identification of biomarkers, which are of great value for both diagnosis and prognosis. Even in cases where metastases of unknown origin are discovered, the AI can help identify the primary tumour.
According to co-researcher Dr Bora Uyar, Flexynesis stands out from existing deep learning tools. Many previous solutions proved to be of limited use, inflexible or difficult to implement. Flexynesis, on the other hand, is designed as a flexible toolkit that can be easily integrated into existing research pipelines. The software is available via platforms such as PyPI, Docker and Bioconda, ensuring accessibility and ease of use.
Deep learning, which uses complex neural networks with hundreds of layers, is ideally suited for analysing the many biological factors that play a role in cancer. Whereas conventional methods often study only one aspect, Flexynesis combines data at the DNA, RNA and protein levels. This makes it possible to map the entire interplay of factors that drive the disease.
From research to practice
The research team emphasises that Flexynesis can also be used by clinical users without in-depth AI knowledge. Its user-friendly design should lower the threshold for implementing multimodal data analysis in hospitals. In countries such as the US, this is already being used more frequently in multidisciplinary tumour boards, but in Germany, the use of multi-omics data is still limited to research programmes such as the MASTER project.
This approach is expected to become more widespread in the coming years, partly thanks to tools such as Flexynesis. This represents a step towards more personalised cancer care, in which therapies are optimally tailored to the unique characteristics of each patient.
AI-driven chemotherapy
Earlier this year, a clinical study conducted by the National University of Singapore showed that AI can make chemotherapy more personalised and effective. Using the AI platform CURATE.AI, researchers were able to accurately tailor doses of the widely used drug capecitabine to individual patients with advanced tumours.
The system uses biomarker data and digital “twins” to predict the optimal dose for each patient. The results show that almost all (97.2%) AI recommendations were adopted by doctors. It was striking that some patients benefited from doses that were on average 20 per cent lower, without any loss of effectiveness. This can lead to fewer side effects and lower healthcare costs.
