Molecular Dynamics and Machine Learning for the Design of Peptide Probes for Biosensing

Human breath and skin surface chemistry contain rich mixtures of molecules and biomarkers that can indicate respiratory infections, diabetes, stress, and other conditions. Fast-acting sensors for human breath could enable rapid diagnosis of diseases like future variants of COVID-19, Respiratory Syncytial Virus (RSV), cancer, diabetes, and more, through detection of unique chemical fingerprints for each type of disease.

Unfortunately, current sensor technologies are too bulky, expensive, or unable to distinguish between different illnesses or conditions. This project will advance the state-of-the-art in biosensing by designing disease-specific sensor arrays that identify combinations of dozens of molecules that comprise unique “fingerprints” found in human breath.

The project will employ physics-based models of molecules interacting with the sensor, state-of-the-art experiments characterizing new sensors, and machine learning (ML) to analyze and design these sensing devices. Data generated from the project will allow the team to explore millions of potential “eNose” designs, creating optimal sensors to detect target diseases.

The team will also use the technical research as a platform to support workforce development and broadening participation in STEM. The work will lead to the training of a PhD student with expertise in biosensing, new materials for classroom instruction in emerging technology, and financial support opportunities for undergraduate summer research experiences.

Molecular dynamics simulations will explore the physical characteristics of peptide-based binders of volatile organic compounds (VOCs). A high-throughput simulation workflow will calculate structure/function relationship for hundreds to thousands of peptide/VOC binding pairs. This data will then be used to develop a sequence-specific ML model that will permit inverse design of new sequences with ideal binding properties.

The most promising molecules will be synthesized and tested using analytical tools and transistor sensor chips for compact eNose systems. This project will synthesize experimental and computational molecular engineering, deep machine learning, and biological sensing mechanisms. By combining experimental data, physics-based simulation, and high throughput ML models, it can, for the first time, assess the true potential sensitivity and specificity of a multi-disease multiplex sensor.

If successful, this project will advance the field and overcome barriers to the fast development of bespoke biosensors for various applications with optimized selectivity and sensitivity. Beyond disease detection, this platform can impact the field of sensing and separations, for example, in the separation of molecules or passive detection for security (e.g., chemical/biological warfare agents).

This work also addresses critical knowledge gaps in existing ML tools for the sequence-level prediction of peptide binders, and the knowledge and data produced in this study will benefit the scientific community and advance the use of these methods broadly in molecular data science. New compact, affordable, noninvasive, and rapid VOC biosensors would create a novel affordable platform for delivering high-demand tests for blood glucose, pregnancy, infectious diseases, and general wellness.

The technology’s future impact could extend to healthcare, public health, agriculture, food storage, environmental monitoring, and defense.

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.