When a woman is diagnosed with breast cancer, the difficult question arises as to which type of treatment is the right one. Gene expression testing is one of the methods used by doctors to help make a prognosis about the course of the disease and, based on this, to select a suitable therapy. However, the reliability of these tests has not been fully established. Scientists from Leipzig University and the Pathologie Hamburg-West institute have now used machine learning to analyse large amounts of data on this question and found that gene expression signatures offer a high degree of certainty in prognosis, but not complete certainty.
Gene expression signatures are descriptions of the activity patterns of genes. When a person is diagnosed with cancer, these signatures can be used to make predictions about how tumours will develop. As such, they are crucial for classifying different types of cancer, determining prognosis and defining treatment strategies.
The current study by Dimitrij Tschodu, a doctoral researcher at the Peter Debye Institute for Soft Matter Physics at Leipzig University, was carried out in close collaboration with Professor Axel Niendorf from the Pathologie Hamburg-West institute and was recently published in the renowned journal Scientific Reports. Tschodu and his colleagues analysed around 10,000 signatures based on renowned breast cancer databases using various machine learning models to thoroughly assess their prognostic ability.
The results of the study show that the gene expression signatures examined lead to a correct patient prognosis in no more than 80 per cent of cases. The researchers also point out that prognoses based on gene expression signatures alone take into account less than 50 per cent of the potentially available information. They therefore recommend using other parameters in addition to gene expression tests. “Although our results confirm the importance of gene expression signatures in predicting patient prognosis, they also highlight the urgent need for a holistic approach that takes into account molecular, clinical, histological and other complementary factors to ensure an accurate prognosis,” explains Tschodu.
Need for a holistic approach to prognosis
“The results of this study are crucial for understanding the limitations of gene expression signatures in cancer prognosis,” adds Professor Josef Käs, head of the Soft Matter Physics Division at Leipzig University. “While gene expression signatures are undoubtedly valuable, our findings show that a holistic approach is needed to ensure an accurate prognosis and to make informed decisions about treatment.”
The publication comes from the Physics of Cancer research field, which looks at cancer from a physical perspective and also examines the mechanics of cells and tissues. Käs says: “This new study underlines the importance of the ‘Physics of Cancer’ in the medical field and the need for interdisciplinary collaboration to find innovative solutions to the challenges in cancer treatment.” Only recently, a research group led by Professor Käs and Professor Niendorf published new findings in this field that could promote more precise diagnostics of the spread and formation of metastases in breast tumours.
Original title of the publication in Scientific Reports: “Re-evaluation of publicly available gene-expression databases using machine-learning yields a maximum prognostic power in breast cancer”, DOI: https://doi.org/10.1038/s41598-023-41090-9
A good work indeed! This translates the difficulty in understanding the behavior of cancer cells to allow development of relevant AI/ML algorithms that can be fed into to diagnose the cancer cells. Relying on a holistic approach is important but should be guarded with some philosophical inquiries.
Congratulations on this marvelous achievement, Prof.
I wish to extend my heartfelt appreciation for the profound contributions made by this research in examining the prognostic utility of gene expression signatures in the context of breast cancer. The study represents a substantial advancement within the realm of cancer research, and its significance cannot be overstated.
This research investigation ventures into the intricacies of utilizing gene expression signatures and machine learning for breast cancer prognosis, constituting a noteworthy endeavor. Notably, it underscores the potential of gene signatures as viable prognostic markers, exhibiting a commendable accuracy rate of approximately 80%. However, it is imperative to acknowledge a key caveat; their effectiveness is not universal, emphasizing the method's inherent limitations. Consequently, the study serves as an important reminder that gene signatures, while promising, should not be regarded as infallible crystal balls. They function more akin to a GPS system, getting you close to your destination but not quite at your doorstep. The study's takeaway message is clear: to attain a comprehensive understanding of breast cancer prognosis, an amalgamation of data elements, including clinical and histological parameters, is indispensable. This multifaceted approach operates much like piecing together a puzzle, where the more components you possess, the clearer the image. Therefore, the study resonates as a reminder of the intricate nature of cancer prognosis and advocates for a holistic toolkit to address the complexities of this critical medical challenge.
Congratulations with your research on such an interesting Oncology field. I know that there are companies who would use microarray technology to predict the impact of a certain therapeutic agent. The health institution, subsequently, can implement therapeutic means more efficiently.
Being active in the gene therapy world, I wonder what your professional opinion is about feasibility to develop gene therapies on these predictive genes?
Regards,
Gökhan
Josef Käs writes articles about himself citing Josef Käs? Most strange!
Concerning AI and cancer there are quite some others around in diagnosis and drug development with lots of investments and highly motivated:
https://www.cbinsights.com/research/ai-startups-fighting-cancer/
E.g. Evaxion Biotech is using AI since 2008.
So what is new here? I am missing a short overview of the state of knowledge.
I just extract that 80% is genetics (so its dominant) and perhaps 20% epigenetics or something else. What is the number for cell mechanics Käs is most interested in? It is less than 20% as we learned, is it less than 10% ?
Cheers!
The maximal concordance index (50% is random and 100% are a safe prediction) that can be achieved by gene expression is 75-80%. 20-25% in the concordance index are missing, which means that 40-50% of all information required for diagnosis are missing. Our recently published breast cancer clinical trial with over a 1000 patients shows that cancer cell unjamming as a new prognostic criteria combined with the Nottingham index as current gold standard leads to a 25% improvement in the concordance index. There is not one smoking gun basic molecular information has to be combined with emergent collective information on the tissue level.
Patients during their life time have many more parameters you can measure. AI is not capable to give you more information you provided for training the AI.
Japanese rarely got cancer when they still ate traditional food.
Why?
- Badges
- Science topic