Huaying Zhang stands next to a microscope and smiles at the camera.

AI4BIO Selects Inaugural Projects for Biomedical Discovery

By Marylee Williams Email Marylee Williams

Heidi Opdyke

好色先生TV鈥檚 Center for AI-Driven Biomedical Research (AI4BIO) chose its first research projects that will use artificial intelligence and machine learning to unlock the complex information inside genomes and cells, paired with automated laboratory platforms that can run and refine experiments at scale.

The four selected pilot projects include researchers from the Ray and Stephanie Lane Computational Biology Department, the Mellon College of Science and the .

鈥淭he long-term vision of AI4BIO is to build virtual cell models that integrate with programmable cloud labs, where AI tools and autonomous experimentation work together to accelerate biomedical research,鈥� said Jian Ma, the Ray and Stephanie Lane Professor of Computational Biology and founder of AI4BIO. 鈥淭hese four projects bring together CMU faculty to produce computational models, reusable datasets and automated workflows, along with early prototypes that help move biological discovery forward.鈥�

AI4BIO develops and applies AI and machine learning methods to understand cellular structure and function across molecular, cellular and tissue scales. The center brings together interdisciplinary expertise from across CMU to build computational methods and experimental workflows that address longstanding challenges in biology relevant to human health and disease.

One of the center's new projects will pair AI-led predictive modeling with automated experiments to better understand how genes and environment shape microbial traits and behaviors. The team will build models of dynamic microbial phenotypes and use active learning to prioritize measurements in automated workflows. Researchers on the project include Joshua Kangas, an associate teaching professor in computational biology and co-director of the master鈥檚 program for automated science; Andrew Bridges, an assistant professor in biological sciences; and Oana Carja, an assistant professor in computational biology.

In another project, researchers Jose Lugo-Martinez, an assistant professor in computational biology, and Huaiying Zhang, the Eberly Family Associate Professor in biological sciences, will use AI-guided experimentation to engineer biomolecular condensates 鈥� compartments inside cells that help organize chemical reactions. The researchers hope that machine learning-guided experimentation will help scientists design these condensates with specific properties. Starting from expert-selected sequences, they will use automated experiments, image analysis and Bayesian optimization to identify condensates with the desired properties.

A third research project will develop automated experimental systems to expand the laboratory and clinical use of the adeno-associated virus (AAV), which is commonly used as a delivery vehicle in gene therapy. Researchers currently face challenges when they use AAV to efficiently experiment with cell-specific targets. Combining laboratory automation and machine learning could improve the process. The team will automate AAV cloning and production, screen barcoded libraries in multiple cell lines, and use machine learning to optimize capsids and regulatory sequences for improved efficiency and cell-type targeting. Andreas Pfenning, an associate professor in computational biology, and , Trustee Professor of , are the researchers on this project.

The final project aims to improve the production of stem cells from skin cells, which is possible but technically challenging. Researchers Christian Cuba-Samaniego, an assistant professor in computational biology, and Newell Washburn, an associate professor in chemistry and biomedical engineering, plan to use robotic systems to systematically vary growing conditions, measure how cells respond to these changes over time, and create a high-quality dataset that links specific conditions to stem cell reprogramming. This dataset will train AI models to learn the rules for reprogramming and help rank the next conditions to test, linking prediction with automated experimentation.