New screening method: molecular fingerprinting from blood plasma
In Munich, efforts are being made to diagnose multiple diseases from a single blood drop using AI on plasma, despite past failures.
Researcher Tarek Eissa: The doctoral student is the first author of the study.
(Image: © Thorsten Naeser / MPQ / Attoworld)
Many people are reluctant to have blood taken, but this is exactly what is needed for standard blood tests - and in sufficient quantities. Vial after vial then fills up at the doctor's. The medical industry reacted accordingly enthusiastically when the start-up Theranos claimed in the early 2010s that it could carry out numerous tests "with just a drop of blood" thanks to a new type of analysis machine called Edison. They had even patented their own mini-vessel called "Nanotainer". It is now clear that it was all just hype and even fraud: the sky-high rise with a billion-dollar valuation was followed by a nasty crash that ended with prison sentences for the founders.
But the idea of carrying out numerous tests with small amounts of blood is still not dead. For example, a team of researchers in Jena is working on a laser-supported procedure to determine a suitable antibiotic for the patient from a single blood sample. According to the researchers at Jena University Hospital, the new optical method only requires two to three and a half hours for a bacterial analysis. Not as fast and much more specialized than the Theranos idea, but reliable in experiments. However, a team at the Ludwig Maximilian University of Munich (LMU) is pursuing a broader approach. The aim there is even to carry out population-wide health screening with just one drop of blood - see Theranos.
Detecting pre-diabetes
A method from infrared spectroscopy is used, known as Fourier transform infrared spectroscopy, or FTIR for short. However, the sample is not diluted, but consists of plasma. LMU has collaborated with the Max Planck Institute of Quantum Optics (MPQ) and the Helmholtz Zentrum München. The FTIR output is read into a computer that has previously learned by means of machine learning (ML) what a "diseased" sample is and which diagnoses belong to it. The LMU team works as part of the Attoworld project and is part of the Broadband Infrared Diagnostics (BIRD) research group. It has already developed a method for creating a kind of molecular fingerprint from plasma. Helmholtz Munich is now adopting the method for a large-scale population study.
Already 5000 samples of blood plasma have been checked using FTIR. This involves determining a correlation between known values, which stand for diseases or diagnoses, with the sample. "A multi-stage computer algorithm is now able to differentiate between various health conditions, including abnormal blood lipid levels, various changes in blood pressure and type 2 diabetes," said LMU in a press release. Surprisingly, this also includes prediabetes, a precursor to diabetes that is otherwise often overlooked by doctors.
Biobanks as a source of data
However, the principle only works if sufficient sample material including suitable findings are available. Only then is it possible to train the ML algorithms accurately enough. In the current project, the LMU team and its colleagues were able to use blood plasma samples from thousands of participants as part of the so-called KORA study, a "comprehensive representative health research project in the Augsburg area". The Helmholtz project is attempting to determine the correlations between health, illness and the living conditions of the population. The focus is on diabetes, cardiovascular and lung diseases as well as environmental issues, according to the DigiMed Bayern platform.
The dream of Theranos founder Elisabeth Holmes, currently incarcerated at Prison Camp Bryan northwest of Houston, appears to be achievable in the foreseeable future: "With a single drop of blood and infrared light, a powerful tool is available to keep an eye on health, identify problems more efficiently and improve healthcare worldwide," says the LMU team. They see "far-reaching potential applications". According to the group, the algorithm was even able to filter out people who were healthy and remained healthy over the study period of several years - and could therefore serve as a positive example to investigate what exactly they were doing right.
(olb)
