Russian Scientists Develop Neural Network Search Tool for Drug Discovery
The new system is designed to identify biologically active compounds and could reduce the time needed to develop innovative medicines.
Researchers at Volgograd State Medical University have developed a neural network that predicts the effectiveness of potential drugs while accounting for complex biological interactions.
The system analyzes the structural formulas of chemical compounds and determines which of them could serve as the basis for treatments targeting psychological or somatic disorders.
The neural network first calculates the strength of interaction between each compound and dozens of biological targets across their full structure, creating a detailed energy profile. A separate algorithm, designed to mimic the functioning of neural networks in the brain, then aggregates these features and converts the dataset into a compact vector metric of affinity. Based on this metric, the system predicts the potential effectiveness of a compound for a specific disease.
Demonstrated Effectiveness Twice
The system’s ability to detect patterns across different areas of pharmacology was tested twice. In the search for potential sedative compounds, it achieved an accuracy rate of 68.3 percent. When identifying compounds active against Staphylococcus aureus, accuracy reached 90.5 percent.
The researchers say the tool provides more than just faster screening for one category of drugs. It offers a flexible platform that can be adapted to a range of diseases.
