Russian specialists from the HSE University have created innovative neural network models for predicting multifactorial diseases. The new method significantly surpasses existing algorithms in determining the risks of developing type 1 diabetes, obesity, and psoriasis.
The development is based on taking into account the complex interaction of genes with each other, which was previously practically not taken into account in medical forecasts. This interaction, known as epistasis, can both enhance and weaken the influence of individual DNA regions on the development of diseases.
The results of our study show new opportunities for personalized medicine and prevention. A more accurate assessment of individual risks will help doctors develop effective strategies for treating and preventing diseases. — notes Maria Poptsova from the International Laboratory of Bioinformatics at HSE University.
Scientists trained the neural network on a database including information about the DNA of more than 58 thousand people of European origin. Special attention was paid to synthetic genomes, where two or three regions interacted with each other, influencing the development of diseases.
Tests showed impressive results — the new recurrent neural network determines the risk of developing type 1 diabetes with an accuracy of 83%, while traditional algorithms cope with this task in only 78% of cases. This discovery opens up new horizons in personalized medicine and therapy.
Read materials on the topic:
Russian researchers have synthesized an antibiotic against resistant Staphylococcus
Dozens of times more effective than chemotherapy: UrFU scientists defeated cancer with a new drug
