At the SPIEF, Skoltech Rector Yulia Gorbunova stated that artificial intelligence will not take away jobs, but will create new ones – provided that specialists learn to work at the intersection of different fields. As an example, she cited the development of young scientists at the institute: they found a way to assess the risk of depression using blood markers.
Skoltech scientists discovered that the severity of depression symptoms is associated with changes in certain lipid molecules in plasma. Based on this data, they created a model that distinguishes people with pronounced symptoms from those without such signs.
Researchers analyzed the blood of 604 city residents and compared the lipid content with the results of questionnaires in which participants assessed signs of anxiety and depression. Separately, they studied samples from 32 patients with clinical depression and 21 healthy individuals. Scientists found eight lipids whose content was associated with the severity of depressive symptoms. Moreover, the changes in volunteers with high scores were similar to the pattern found in patients with a confirmed diagnosis.
The work is part of a larger Skoltech project to create a method for objective assessment of the risk of mental disorders based on blood lipids. The research is being conducted jointly with the N. A. Alekseev Psychiatric Clinical Hospital No. 1 and the Moscow Center for Innovative Technologies in Healthcare. The team of scientists is also investigating whether such markers can distinguish depressive disorders from schizophrenia.
In a later study, scientists examined the blood of 416 patients with mental disorders and 272 healthy individuals from Moscow and Ufa. They found 107 lipid molecules that changed in both depression and schizophrenia, and another 37 that changed only in one of these conditions. Based on this data, the algorithm learned to distinguish major depressive disorder and schizophrenia with an accuracy of about 83%. In addition, Skoltech reported that the early risk assessment system showed an accuracy of 93% on a general sample of almost 5,000 people.