Machine Learning Uncovers 84 Novel Cancer Risk Factors in UK Biobank Study
Machine learning has played a significant role in a recent UK Biobank study, revealing 84 previously unknown features that could indicate an increased risk of cancer. The research, titled Hypothesis-free discovery of novel cancer predictors using machine learning, was conducted by a team of researchers from the University of South Australia (UniSA) and the University of Adelaide. Their groundbreaking approach involved deploying artificial intelligence and statistical methods to analyze data from over 459,000 participants in the UK Biobank.
Dr. Iqbal Madakkatel, one of the researchers involved in the study, explained the significance of their hypothesis-free analysis, highlighting the use of AI to identify cancer risk factors among thousands of features. The researchers discovered that over 40% of the identified features were biomarkers, biological molecules that can indicate health or unhealthy conditions. Intriguingly, several of these biomarkers were found to be associated not only with cancer risk but also chronic kidney or liver diseases.
One of the key findings of the study was that high levels of urinary microalbumin, a protein found in urine, were the strongest predictor of cancer risk after age. While microalbumin is crucial for tissue growth and healing, its presence in urine can serve as an indicator of kidney disease and also signal an escalated risk of cancer. The researchers also observed that elevated levels of other markers associated with poor kidney performance, such as cystatin C and urinary creatinine, as well as lower total serum protein, were linked to an increased risk of cancer.
Interestingly, the study also revealed that a higher red cell distribution width (RDW), which indicates variations in the size of red blood cells, is tied to a heightened risk of cancer. Normally, red blood cells should have a consistent size, but when discrepancies occur, it can be correlated with inflammation and poorer renal function, ultimately leading to a higher risk of cancer.
C-reactive protein, an indicator of systemic inflammation, and the enzyme gamma glutamyl transferase (GGT), a biomarker for liver stress, were also found to be connected to an increased risk of cancer. These discoveries shed light on the underlying pathogenic mechanisms of various diseases, including cancer, kidney disease, and liver disease, emphasizing the importance of further exploring their potential connections.
Professor Elina Hyppönen, the chief investigator and the Centre Director of the Australian Centre for Precision Health at UniSA, emphasized the strength of their study lies in the power of machine learning. By incorporating and cross-referencing thousands of features, including clinical, behavioral, and social factors, their model can uncover relevant risk predictors that might otherwise remain hidden. Although further studies are required to confirm causality and clinical relevance, this research suggests that simple blood tests could provide valuable information about individuals’ future risk of cancer, enabling early interventions to potentially prevent the disease.
In conclusion, the utilization of machine learning in the UK Biobank study has led to groundbreaking discoveries of 84 novel cancer risk factors. This innovative approach has not only revealed potential indicators of increased cancer risk but has also provided insights into the connections between cancer and other diseases. The findings open up new avenues for early detection and prevention strategies, potentially revolutionizing the field of cancer research.
