Physics-informed machine learning (PIML) is emerging as a powerful framework for modeling complex biological systems by integrating mechanistic knowledge with experimental data. Unlike purely data-driven approaches, PIML incorporates differential equations describing biological processes while enabling unknown mechanisms to be learned directly from sparse, noisy, and partially observed measurements. This capability is particularly valuable in biology, where mechanistic models are often incomplete and experimental data are limited. In this talk, I will present recent advances from our group in developing interpretable physics-informed learning methods for biomedical applications. I will introduce Universal Physics-Informed Neural Networks (UPINNs), which extend conventional PINNs by simultaneously estimating unknown parameters and discovering previously unknown biological functions from data. I will demonstrate applications in quantitative systems pharmacology, drug development, and cancer biology, and discuss how these approaches can accelerate mechanistic discovery, improve predictive modeling, and support the development of next-generation biological digital twins.