KAIST GenAI model predicts unseen drug-cell reactions, guiding cancer and regenerative treatments with precision.
Researchers at KAIST have developed a GenAI system that predicts how cells respond to drugs and genetic changes. This can potentially enhance drug discovery and personalized medicine. Led by Professor Kwang-Hyun Cho, the team created an AI model that simulates cell–drug interactions like Lego blocks. They break them into modular parts and recombining them to forecast new, untested reactions.
One of the biggest challenges in life sciences is controlling how a cell behaves. This is essential for cancer therapy, genetic regulation, and regenerative medicine. Traditional methods rely on costly lab experiments and limited datasets. KAIST’s GenAI model overcomes these barriers by mapping the latent space of cell states and drug effects. It separates, then reassembles these representations to predict what would happen if a new drug or gene intervention were applied.
Using real-world experimental data, the AI accurately predicted molecular targets that could revert colorectal cancer cells to a healthy state—later verified in laboratory experiments. This proves the system’s ability to identify both effective drugs and their internal mechanisms. Beyond oncology, the technology could aid in regenerating damaged tissues and designing new treatments by simulating cell transitions before physical testing.
Professor Cho explained that the inspiration came from image-generation AI, applying the idea of “direction vectors” to cellular biology. This allows the AI to model how cells move toward desired states under different treatments. The result is a universal, quantitative AI platform that predicts how genes or drugs reshape cellular behavior—paving the way for faster, safer, and more precise biomedical innovation.
