Big data in medicine: advancements in diagnosis and treatment

A look at the digital medicine of the future

"Modern molecular medicine alone witnessed more data generated in 2015 than in the entire period from 1990 to 2005," explains Burkhard Rost, Professor of Bioinformatics at the TUM. "And this is going to continue at this rate." However, up to now, when it comes to preparing, analyzing and applying this treasure trove of data, we have lagged far behind the available technical possibilities.

We also lack the necessary algorithms and the ability to link diverse areas such as medicine and biology, on the one hand, and computer science, on the other. "A biologist cannot interpret the gigantic quantities of data on his or her own," stresses Hans-Werner Mewes, Professor of Genome-Oriented Bioinformatics at the TUM. "Here you need bioinformatical methods."

At the TUM, a whole host of scientists from different disciplines have set themselves a target of collecting the data treasure trove from the life sciences and making it usable for researchers, patients, doctors and clinics. In addition, specialists are taking care of data security. Together with the mainframes at the Leibniz Supercomputing Center in Garching, the university has a unique infrastructure for bioinformatics and medical informatics.

Crowdsourcing – data analysis as a computer game

Medical data is still in a pretty chaotic state today: laboratory data pertaining to blood or histological examinations, biochemical information from genome sequencing and from endogenous protein building blocks (proteome), statistical results from numerous experiments and clinical studies, and graphical data. However, this means that big data in this form is not accessible for practical use. That is why researchers are looking for ways to organize this flood of information and bundle it into uniform databases. You can explain and add to this information by providing annotations that tell you more about the origin, properties and content of the data.

At Professor Nassir Navab’s Chair for Computer Aided Medical Procedures & Augmented Reality, researchers are working toward simplifying the assignment of such annotations, among other things. "Doctors and laboratory staff rarely have time to take care of processing their data," explains Shadi Albarqouni. He has therefore developed a project, which enables large numbers of Internet users to make contributions. This process is called crowdsourcing and it allows anyone who has the time and the motivation to participate in the completion of certain digital tasks. In Albarqouni’s case, this means analyzing histological tissue sections for breast cancer, for example.

"You have to tell the computer which cells from the image are cancer cells, so that it can learn and ultimately recognize cancer cells by itself," explains the researcher. "This is in principle a really tedious task and usually not all of these cells are captured in a histological section due to time constraints.” To get Internet users with little experience involved in such evaluations as well, he has developed a computer game, in which the aim is to "shoot" as many malignant cancer cells as possible. "People have fun in this gaming environment and target only those objects that particularly resemble cancer cells." The computer registers the cells that have been "shot" and gives them the annotation "cancer cell" internally.

Complicated manual of life

Whether through games or not, the objective remains the same - to transform data material into usable information. How do you arrange and organize genome and proteome data, for example? "When the sequencing of the human genome was completed in 2001, everyone rejoiced, believing that the manual of life had been discovered," explains Burkhard Rost. "However, as with all manuals, it is also the case here that you cannot find what you actually need and if you do find it, you do not understand it. Today, as we interpret the genetic sequence, we have only pushed open a tiny window in the cathedral of knowledge."

For example, the prevailing view for a long time held that there was very little difference between the genes of healthy people. That is, the sequencing of an individual genome would be enough to cover almost all variants. "We now know today that that’s wrong," explains Rost who transferred to the TUM from Columbia University, New York in 2009. "Let’s look at the proteins in our body. We have approximately 20,000 variants, about a quarter of which differ between two people in the most essential of terms."

Therapy success thanks to precision medicine

This can impact, for example, on how and whether chemotherapy works. Some tumors are insensitive to certain treatments or the active ingredient is rendered harmless by the cancer cells. "For the majority of mass tumors such as breast, lung, stomach and bowel cancer, only around a fifth of patients respond well to the conventional chemotherapies," explains Professor Bernhard Wolf, incumbent of the Heinz Nixdorf Chair for Medical Electronics at the TUM. "To avoid putting patients under unnecessary strain, and to increase therapy success rates and thus reduce costs in the long term as well, it is therefore imperative that we personalize the therapy."

Since, up to now, it has not been possible to determine the sensitivity of patients using genetic markers with any degree of reliability, a team from Wolf’s chair has developed sensors that measure in advance how strongly cells, which are taken from the patient, react to various chemotherapeutic agents. Based on this effect, the doctor can then apply the active ingredient that is best suited to the individual case. The system is ready for use and is currently being tested in a preclinical study in cooperation with the Asklepios Clinic Barmbek in Hamburg. However, ideally, there would be a gene test for the patient to determine the correct chemotherapeutic agent automatically.

The ideal scenario would be precision medicine. This means aiming to have the focus on the individual characteristics of the patient, so that predictions, therapies and prognoses can be tailored precisely to the needs of the individual. In the long term, the doctor should be able to know which treatment best suits the patient based on his or her genetic profile. "The individual genomic profile is then noted on the health card, meaning that each patient receives precisely the medication that he or she reacts well to," according to Rost's vision of the future. "Even a person’s diet could be optimally tuned to the genetic properties of the individual."

On the trail of rare diseases

However, there is still a long way to go yet. This is due to a second major error, namely the idea that the non-coding DNA fragments that are located between the known genes serve no purpose at all; this so-called "junk DNA" as it has been labeled. Researchers gradually discovered that parts of this DNA, still amounting to around 95 percent of the human genetic material, also perform important functions; for example, these parts deal with the turning on and off of genes and also contain information about the evolutionary development of the organism.

Professor Mewes gives the example of research work that examines who is more or less likely to become obese. "The regulation of the gene plays an important role here," he explains. "Therefore, it is difficult to understand why there are flaws within the metabolism, even if genetic population studies give us an indication." He and his colleagues specialize in deciphering and interpreting such data with bioinformatical procedures - for example, using neuronal networks - and with the technology of machine learning.

By analyzing naturally occurring genetic variations in the proteins of the human body, both Hans-Werner Mewes and Burkhard Rost want to find out, for example, which mutations are responsible for rare diseases. These are diseases that are rare if taken in isolation but affect about five percent of the population in total. They can surface very early in life - even before birth in many cases - and can accompany a person throughout their life. "They are almost impossible to detect using traditional methods; you usually need five to 20 years before getting a clinical diagnosis," explains Hans-Werner Mewes. "However, if the genetic defect is found, you know in 25 to 40 percent of cases exactly whether therapeutic measures can specifically be taken."

On the search for risk genes for schizophrenia and cancer

A current example from the research field is the search for genetic variants that are associated with schizophrenia. It is one of the most severe mental illnesses, affecting 0.5 to one percent of the global population, with about half of those affected suffering their whole life. Numerous family studies show a strong genetic component. However, up to now, there has been no catalog of mutations that cause schizophrenia. "The talk among experts is about whether rare genetic variations are responsible for the occurrence of schizophrenia," says Mewes. "Clarifying the causal relationships is important for early diagnosis and treatment, as well as for the development of new therapies."

This also applies to cancer research. Bioinformatical procedures are also essential here. You can use these procedures to examine and compare large quantities of data; and indications of dangerous mutations can already be found in this way through mathematical methods alone - fully independent of causal investigation initially. Furthermore, you can compare genetic patterns of a cancer patient with those of many others in a database and find out where the largest correlations are. In this way, similar to Amazon’s approach with book recommendations, you can get a greater understanding without having to make assumptions about mechanisms of action.

Intelligent sensors - virtual connection to the doctor

Digitalized medical data is not only used for research, but it can also help to ease the pressure on doctors directly if they can be put in virtual contact with chronic patients, in particular. In recent years, Bernhard Wolf and his team have developed a prototype for this with the COMES system. It uses various sensors to collect individual data and transmits it to a database. This data is processed and evaluated there. If a value exceeds the predefined threshold values, the system automatically alerts the attending doctor or the responsible nursing service on their cell phone.

In this way, pulse and blood pressure can be monitored and you can also determine from the skin resistance whether the person is adequately hydrated. Similarly, the device can transfer blood sugar levels and monitor weight. What is always important here is that the patient must release the data; that is, he or she is not forcibly monitored. Because COMES is designed in such a way that, on the one hand, users can check their medical data using a database and release it to the doctor when required; on the other hand, they receive additional information themselves via feedback systems.

Doctors, meanwhile, can track patients equipped with COMES in the therapy, intervene by way of precaution based on the relevant data or initiate interventions via the medical center in the event of absence or inability to attend. "We have determined in numerous studies that the regular sensory support makes patients feel safer and yet more independent," says Professor Wolf. A spin-off is now set to bring the device to market. In practical trials, COMES has already produced some interesting practical results. For example, researchers were able to show that certain pieces of music or sound patterns have the effect of reducing blood pressure in many patients.

Another example is the intelligent tooth splint called SensoBite. This is a dental splint, as prescribed by a dentist in the event of bruxism during sleep. A piezoelectric sensor system is integrated into this splint to measure the masticatory movements. A radio transmitter sends the measured data wirelessly to a receiver that is the size of a matchbox and is located on the body of the patient or where he or she sleeps. A USB interface can be used to transfer the stored data to the computer of the attending doctor or a vibration signal can be sent to the person sleeping via biofeedback. In this way, doctors can analyze the causes of the grinding or patients can break the grinding habit directly themselves.

Medical informatics: efficient use of medical data

The flood of data in the medical sector is increasing day by day. "In a few years, we will know the individual genomic data of many millions of people," says Hans-Werner Mewes. "We will not be able to use this data without efficient procedures in place in bioinformatics and medical informatics." However, the data still remains inadequate for many purposes. "Today doctors only ever consider their own patients," according to the researcher. "However, it would be important during the stage of evaluating medication and therapies if you had comparative data from as many patients as possible over the course, because factors such as age, metabolism and genetics all contribute to the treatment outcome. For this to work, you would have to install systems that record and evaluate this type of information, moving away from traditional studies and toward day-to-day practice."