Design-by-Learning and Learning-from-Design: White-Box Data-Driven Design of 2D Material-Based Biosensors and Its Impact on Designers

About This Grant

This project investigates how artificial intelligence can overcome current challenges in designing biosensors based on two-dimensional materials, moving beyond laborious trial-and-error methods, while simultaneously enabling human experts to acquire new knowledge in this domain. Rational design for these crucial devices faces four primary challenges: insufficient data due to expensive experimentation and inherent measurement uncertainties; the absence of accurate theoretical models to capture complex biochemical interactions; the constant demand for rapid responses to emerging needs (e.g., new pathogens); and the necessity for transparent, data-driven design approaches in high-stakes biomedical applications. To address these issues, we propose a “white-box” data-driven design and knowledge discovery framework that integrates interpretable machine learning, data fusion, and statistical inference. This methodology seeks to accelerate on-demand biosensor development through a “design-by-learning” paradigm and enhance expert capabilities to address new demands by providing interpretable insights derived from the design process (“learning-from-design”). This project looks to advance biosensor design, promote national healthcare readiness, and support STEM education through novel research and practical applications. The developed design methodologies are expected to generalize beyond biosensor development to broader areas, including advanced materials design. This research looks to spearhead studies of AI trustworthiness in data-driven design for high-stakes applications, shedding light on how model transparency impacts design cognition and performance. The core objective of this project is to establish a “white-box” data-driven design framework for 2D material-based biosensors, emphasizing model transparency and quantifying its influence on designers’ knowledge acquisition, perception, and overall performance. To achieve this, the research looks to develop and implement several key innovations: (i) the development of novel methods for discovering analytical relationships between biosensor properties and performance, even when confronted with highly uncertain experimental data; (ii) the creation of interpretable multi-fidelity modeling approaches designed to distinguish between generalizable and non-generalizable influences of design variables, thereby guiding reliable extrapolation to new design scenarios; and (iii) design studies to quantify how model transparency impacts designers’ cognitive processes, knowledge acquisition, and performance, thereby advancing our understanding of the practical importance of transparency in data-driven design. This research framework will undergo rigorous evaluation through its practical application in designing biosensors for various biomarkers. The developed design interfaces will be showcased in dedicated workshops aimed at disseminating findings, training new users, and gathering valuable feedback to quantify the impact of this research and further refine the framework. 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.