Anything medically related these days seems to have to mention patient data, algorithms or machine learning. Are these terms just the latest trendy meditech-jargon, or do they actually relate to something more meaningful in practice?
Medicine largely relies on the ability to come up with a correct diagnosis. The diagnosis is evidently absolutely critical. If the diagnosis is not right, then the recommended treatment is unlikely to help, and could indeed cause harm. Doctors normally arrive at a diagnosis by considering all the symptoms, which may also be further informed by clinical laboratory testing (pathology tests). So what's the problem?
The trouble is that there are many types of disease, or indeed sub-types of diseases (pathologies) that cause very similar symptoms. Furthermore, routine laboratory testing, even with the new PCR and molecular testing has its own limitations and is sometimes unable to assist in confirming the correct diagnosis. In any event, most diagnostic tests are not absolutes and compare your result to a 'normal population range'. This point is problematic in its own right as 'normal ranges' can be quite wide and even if you fall outside of a normal range for a test, it does not necessarily mean it is a sign of anything untoward, as your healthy 'normal' may just be different.
So, something more is required in order to provide the level of diagnostic discernment that is needed in these cases, and that's where data 'clustering' and the algorithms come into play. However, making sense of these often complex data requires automated web tools or other software to do the complex analysis.
Why do we use algorithms for medical diagnosis?
Different causes can cause the same symptoms. Algorithms help to distinguish diseases at the molecular level.
Machine learning is playing an ever-increasing role in biomedical research. Scientists at the Technical University of Munich (TUM) have now developed a new method of using molecular data to extract subtypes of illnesses.
a web-based tool that permits online analysis of molecular clinical data by practitioners without prior knowledge of bioinformatics
Doctors define and diagnose most diseases on the basis of symptoms. However, that does not necessarily mean that the illnesses of patients with similar symptoms will have identical causes or demonstrate the same molecular changes. In 'biomedicine', one often speaks of the molecular mechanisms of a disease. This refers to changes in the regulation of genes, proteins or metabolic pathways at the onset of illness.
The goal of stratified medicine is to classify patients into various subtypes at the molecular level in order to provide more targeted treatments.
To extract disease subtypes from large pools of patient data, new machine learning algorithms can help. They are designed to independently recognize patterns and correlations in extensive clinical measurements. The LipiTUM junior research group, headed by Dr. Josch Konstantin Pauling of the Chair for Experimental Bioinformatics has developed an algorithm for this purpose.
Analysing complex data with an automated web tool
Their method combines the results of existing algorithms to obtain more precise and robust predictions of clinical subtypes. This unifies the characteristics and advantages of each algorithm and eliminates their time-consuming adjustment. “This makes it much easier to apply the analysis in clinical research,” reports Dr. Pauling. “For that reason, we have developed a web-based tool that permits online analysis of molecular clinical data by practitioners without prior knowledge of bioinformatics.”
Biclustering technology and algorithms
On the TUM website, researchers can submit their data for automated analysis and use the results to interpret their studies. “Another important aspect for us was the visualization of the results. Previous approaches were not capable of generating intuitive visualizations of relationships between patient groups, clinical factors and molecular signatures. This will change with the web-based visualization produced by our MoSBi tool,” says Tim Rose, a scientist at the TUM School of Life Sciences. MoSBi stands for “Molecular Signatures using Biclustering”. “Biclustering” is the name of the technology used by the algorithm.
How are using these algorithms relevant in practice?
With the tool, researchers can now, for example, represent data from cancer studies and simulations for various scenarios. They have already demonstrated the potential of their method in a large-scale clinical study. In a cooperative study conducted with researchers from the Max Planck Institute in Dresden, the Technical University of Dresden and the Kiel University Clinic, they studied the change in lipid metabolism in the liver of patients with non-alcoholic fatty liver disease (NAFLD).
The exact causes of NAFLD aren’t well understood. There appears to be a connection between the disease and insulin resistance.
Broken veins are a surprising symptom, however for some there will be no sign that this problem is occurring.
Other symptoms of non-alcoholic fatty liver disease include:
- Yellowing of the skin and eyes
- Swollen glands
- Discolouration in the neck
- Swelling in the belly
- Pain in the upper right side of the abdomen
A healthy liver should contain little or no fat. It's estimated up to 1 in every 3 people in the UK has early stages of NAFLD, where there are small amounts of fat in their liver.
From NAFL to NASH
This widespread disease is associated with obesity and diabetes. It develops from the non-alcoholic fatty liver (NAFL), in which lipids are deposited in liver cells, to non-alcoholic steatohepatitis (NASH), in which the liver becomes further inflamed, to liver cirrhosis and the formation of tumors. Apart from dietary adjustments, no treatments have been found to date. Because the disease is characterized and diagnosed by the accumulation of various lipids in the liver, it is important to understand their molecular composition.
Discerning the differences in biomarkers for liver disease
Using the MoSBi methods, the researchers were able to demonstrate the heterogeneity of the livers of patients in the NAFL stage at the molecular level. “From a molecular standpoint, the liver cells of many NAFL patients were almost identical to those of NASH patients, while others were still largely similar to healthy patients. We could also confirm our predictions using clinical data,” says Dr. Pauling. “We were then able to identify two potential lipid biomarkers for disease progression.” This is important for early recognition of the disease and its progression and the development of targeted treatments.
The research group is already working on further applications of their method to gain a better understanding of other diseases. “In the future algorithms will play an even greater role in biomedical research than they already do today. They can make it significantly easier to detect complex mechanisms and find more targeted treatment approaches,” says Dr. Pauling.