Systematic Closed-Loop Identification Method for Volatile Organic Compound-Selective Peptide Recognition Element on Carbon Nanotubes

Ubiquitous, autonomous, and real-time biomarker sensing is imperative to human health and performance monitoring. Volatile organic compounds (VOCs) found in exhaled breath are essential biomarkers for assessing human physical and physiological status. Biorecognition elements (BREs), biological materials with high affinity to the target molecules, can be used to detect VOCs. Among BREs, short peptides are promising for realizing sensors as they offer chemical stability and design flexibility. Although high-throughput methods are available to identify potential peptide libraries using computation and biological assays, realizing the tangible device requires follow-up experimental characterization in an operationally relevant environment to optimize peptide-based sensors. However, this <i>in-operando</i> characterization is challenging and time-consuming because the interface between peptide and tangible devices is an additional factor that can compromise functionality. Here, we present a novel closed-loop pipeline to optimize peptide-functionalized sensors <i>in-operando</i> based on data-driven feedback. As a proof of principle, a carbon nanotube (CNT) chemiresistor platform serves as a tangible device. CNTs have been widely used as chemical sensors due to their advantageous characteristics, such as being nano-sized, mechanical/electrical stability, and compatibility with functional materials. The suggested method comprises library identification, solution preparation, fabrication, output acquisition, and analysis, and then data-driven feedback suggests new peptide combinations for the specific targets to iterate the following pipeline to reach improved performance. First, the pipeline randomly selects two peptides from the library. Then, each step is performed and evaluated regarding the solubility of the peptide in the CNT solution and device fabrication yield. The fabricated devices generate outputs that are CNT resistance changes in response to the VOC. The output data are used to perform cluster analysis to evaluate the performance in differentiating target VOCs. The analysis results and peptide properties provide an idea to suggest a new combination of peptides, and then the subsequent pipeline starts testing the suggested peptides. These feedback-based systems improved the clustering performance of the peptides after three iterations, demonstrating efficient and rapid optimization compared to random repetition. This methodology paves the way for the high-throughput design of tangible devices functionalized by the peptide sequence specific to target VOCs and developing susceptible CNT-based VOC sensors.