CAREER: Machine Learning Algorithms for Tackling Annotation Inequality in Protein Function Characterization

Much of life science research is centered on understanding the functional roles of proteins. However, most scientific attention has historically focused on a limited set of increasingly well-known proteins, while the biological functions of the vast majority remain largely unknown, leading to an “annotation inequality” that biases our understanding of protein functions. This project aims to advance protein function annotation by developing artificial intelligence (AI) and machine learning (ML) methods to improve the accuracy and coverage of protein function predictions and to bridge the gap in function knowledge between understudied and well-characterized proteins. The research has potential impacts on advancing our understanding of various topics centered around protein biology and yields new analytic algorithms with broad basic biological and biomedical applications, such as drug discovery, vaccine development, and personalized diagnosis and treatment, ultimately contributing to improvements in human health as well as understanding of basic biology. The research activities are tightly integrated with education and outreach efforts. This project introduces a systematic computational framework for protein function annotation. The research focuses on developing new ML models that are specifically designed to tackle core challenges in protein function annotation, such as annotation bias, data sparsity, and the need for heterogeneous data integration and error-controlled prediction. It emphasizes a biology-first approach, building ML models that incorporate biological principles and data uncertainty. The project builds on and advances several areas of bioinformatics and ML—including system biology, structural and functional genomics, multi-modal representation learning, generative modeling, uncertainty quantification, and mathematical optimization—to create a unified platform for accurate, scalable, and unbiased protein function prediction. The resulting data, models, and software will be made publicly available to the broader research community to support new data-driven scientific discoveries and practical applications in computational biology, basic biology, and biomedicine. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.