A recent publication in Biology Methods & Protocols by Oxford University Press presents a promising future where artificial intelligence (AI) may assist physicians in detecting and diagnosing cancer, potentially leading to earlier interventions. Cancer presents a significant challenge annually, with over 19 million new cases and 10 million deaths. The evolutionary complexity of cancer often renders late-stage tumours challenging to manage, but the potential of AI offers a ray of hope.
DNA stores genetic data using sequences of four bases—A, T, G, and C. External environmental factors can lead to alterations in these bases through a process known as “DNA methylation,” where a methyl group is added to the DNA. This modification occurs extensively across the millions of DNA bases in each cell. Notably, early stages of cancer are marked by distinct changes in these methylation patterns, which could be crucial for early cancer detection. The task is to determine the methylation statuses across cancerous DNA and compare these to normal tissue, a daunting task likened to finding a needle in a haystack.
In an exciting development, Cambridge University and Imperial College London researchers have developed an AI model that employs machine learning and deep learning techniques to scrutinize DNA methylation patterns. This model can distinguish between 13 different types of cancer—including breast, liver, lung, and prostate cancer—and non-cancerous tissue with an impressive 98.2% accuracy. However, this model currently analyzes tissue samples rather than DNA fragments in blood, indicating a need for further training on a broader array of biopsy samples to prepare them for clinical application.
This study’s use of an explainable and interpretable AI core is particularly significant. It allows for a deeper understanding of the model’s decision-making processes, enhancing knowledge of the biological mechanisms involved in cancer. This detailed examination of the model’s functioning could significantly contribute to cancer research.
Detecting these aberrant methylation patterns, potentially from biopsies, could enable healthcare providers to identify cancer at its inception. This improves patient prognosis as most cancers are more manageable or curable when caught early.
The lead author of the study, Shamith Samarajiwa, underscored the potential of computational methods to revolutionize early cancer detection and screening. With continued improvement and rigorous clinical testing, AI models could soon become invaluable tools in oncology, significantly enhancing patient outcomes through early detection and providing reassurance about the future of cancer treatment.
More information: Izzy Newsham et al, Early detection and diagnosis of cancer with interpretable machine learning to uncover cancer-specific DNA methylation patterns, Biology Methods and Protocols. DOI: 10.1093/biomethods/bpae028
Journal information: Biology Methods and Protocols Provided by Oxford University Press USA
