Artificial intelligence or deep learning has incredible potential for medical discovery. AI can help with difficult diagnoses, identify drugs for hard-to-treat conditions, and much more. With its basis in genomics, the Lieber Institute for Brain Development is uniquely positioned with the experienced data scientists necessary to put AI to good use in accelerating medical research. Here, we sit down with AI expert and Lieber Institute Lead Investigator Dr. Shizhong Han to talk about the future of deep learning in medicine.
What is exciting about the idea of using AI for medicine?
What excites me most is the vast amount of data that the field of medicine has accumulated — from medical images and genomic data to electronic health records. This wealth of data presents an unprecedented opportunity for AI to learn and uncover complex patterns that traditional methods may overlook. This could help us predict disease at much earlier stages, leading to better diagnoses and faster and more personalized treatments. I’m particularly enthusiastic about AI’s potential to accelerate drug discovery. Generative AI can design drug candidates, greatly reducing the time and cost required to develop new therapies.
How is the Lieber Institute using AI for research?
My team at LIBD has applied AI to understand the functional impact of genetic variations, particularly how changes in DNA sequence influence DNA methylation levels, an epigenetic mark that plays a critical role in brain development and function. Epigenetic changes are chemical modifications of DNA that can alter gene function without changing the underlying DNA sequence. We further used these predicted functions of genetic variations to narrow down the genes that cause different brain disorders, with the goal of finding new ways to treat mental illnesses. Our current focus has two key directions: one involves developing single-cell foundation models to zoom in and better understand the underpinnings of human health and disease at the molecular level. The second focuses on leveraging generative AI to design completely new drugs to treat brain disease. By combining these two research lines, we hope to bring science closer to curing and preventing brain disorders.
What’s unique about schizophrenia that makes it a good candidate for AI?
Schizophrenia is a complex disorder with a strong genetic basis, involving potentially hundreds of interconnected risk genes. This complexity makes AI an ideal tool to tackle the puzzle by identifying hidden patterns that traditional statistical methods may struggle to detect. We are now able to create single-cell foundation models or deep-learning models that have trained on big data that describe hundreds of millions of individual cells. By combining these models with unique single-cell datasets from schizophrenia patients, AI has the potential to reveal new clues about what causes disease and help us find new aspects of the brain to target for drug development.
What do you expect to see with AI in medicine in the next five years? What about 10 years?
Over the next five years, I expect AI to become integrated into many aspects of medicine. We will see many AI-generated molecules entering clinical trials, and AI agents will continue to mature and gain broader adoption in both research and clinical practice. AI agents are tools that can execute tasks on their own without human intervention. This may include predictive models for patient outcomes, automated image analysis systems, and personalized treatment recommendation engines.
Looking further ahead to the next 10 years, I envision FDA-approved AI-generated molecules becoming available and AI agents playing a prominent role in both clinical care and research, significantly accelerating progress in these areas. Additionally, AI-driven virtual health assistants may emerge, providing continuous health monitoring and enabling early intervention for chronic conditions.