Algae Against Cancer: How AI Is Helping Researchers Find New Drug Candidates
Russian researchers have developed a faster way to identify compounds in microalgae and cyanobacteria that could potentially be used to fight cancer. They are using a neural network to help with the search.

Microalgae and cyanobacteria are tiny organisms that could hold significant promise for medicine. They contain biologically active compounds with a range of properties, including antioxidant, antibacterial, antiviral and antitumor activity. Scientists have already isolated more than 1,000 such compounds, but the potential of these organisms extends much further.
Estimates suggest that between 200,000 and 800,000 species of microalgae and cyanobacteria exist on Earth. Finding those that could help target tumor cells is no easy task. Researchers at the Moscow Institute of Physics and Technology (MIPT) are working on the problem with colleagues from Sirius University, the Kurchatov Institute, the A.I. Burnazyan Federal Medical Biophysical Center, Oncobox and other research organizations.
Under the proposed approach, artificial intelligence analyzes how tumor cells respond to potentially useful compounds and helps identify promising candidates. This can significantly accelerate the search while reducing the cost of experiments.

AI Analyzes How Cells Respond
The approach is based on functional screening. First, researchers expose tumor cells to extracts from biological samples. They then examine changes in gene activity. These changes can be used to reconstruct the molecular processes triggered by the substance under investigation.
The researchers use the Oncobox platform, developed by Oncobox in collaboration with its partners, to analyze the data. It helps identify which molecular mechanisms have been affected and assess the potential of each sample. The resulting predictions are then compared with the results of direct cell-viability tests.
“Our approach is to first expose HL-60 leukemia cells to microalgae extracts, then examine how gene activity changes and use those changes to reconstruct which molecular processes in the cell have been affected. This allows us to decide which extracts are worth investigating further,” said Denis Kuzmin, director of MIPT’s School of Biological and Medical Physics.

What the Experiment Revealed
The researchers have tested 48 extracts so far: 28 from microalgal strains and 20 from cyanobacterial strains. Conventional laboratory tests found that eight samples could damage cancer cells. The system’s predictions, based on analysis of molecular processes, agreed with the direct test results in 92% of cases, with no false positives. The researchers also detected antitumor activity for the first time in three strains of desmid algae. In their view, this finding broadens the range of candidates available for further experiments.
The researchers are also interested in how the compounds work. They took a closer look at the most active sample, an extract from the cyanobacterium Nostoc sp. SBV-48, because it affected the activity of 290 molecular pathways, including the activation of 25 pathways associated with cell death. Chemical analysis showed that most of the activity was attributable to Cryptophycin-1, which accounted for just 0.01% of the sample by mass. This example also illustrates how difficult it can be to identify active compounds in natural mixtures: even a minute concentration can have a substantial effect on the outcome.
The Search for New Drugs
Russian researchers are steadily advancing tools for discovering cancer treatments. In 2024, the Moscow Institute of Physics and Technology developed a cell-based assay to test how living cells respond to drugs and introduced an automated method for counting fluorescently stained cells, improving the accuracy of the assessments. That same year, researchers at the Higher School of Economics improved GFlowNets neural networks for computational searches for molecules with specific properties. In 2026, Immanuel Kant Baltic Federal University proposed a more efficient method for extracting fucoidan from brown algae and investigated the processing of the red microalga Porphyridium.
The new development by MIPT and its partners builds on this research, bringing the development of more effective cancer drugs a step closer.

A Neural Network Could Save Valuable Time
The main practical advantage of the Oncobox platform is that it can screen out unsuitable samples at an early stage, allowing researchers to focus on the most promising candidates. The more accurately researchers can identify promising compounds, the better their chances of finding substances with effective antitumor properties.
But experts caution that developing a drug capable of treating all forms of cancer remains a distant prospect. So far, the method has been tested only on HL-60 leukemia cells, meaning the findings cannot automatically be applied to other types of tumors.
The researchers’ next task is to test promising samples on other tumor models and on healthy cells. Even at this stage, however, the new approach makes a labor-intensive part of the research process more targeted and systematic, helping scientists identify which samples are genuinely worth investing resources in.
The project’s success could open up new opportunities for Russia’s bioinformatics software market, an emerging field at the intersection of science, medical practice and digital technologies.









































