An Automated High-Throughput Platform for AI-Guided Development of Nanopore Membrane Biosensors and Real-time Edge Processing

Abstract: Biorecognition elements immobilized on nanopore membranes have emerged as a biosensing technique that has been transformative for Point-of-Care (POC) Diagnostics and translational medicine. This is due to their utility as potent physical transducers to convert molecular interactions into electrical and fluorescent signals. However, there is significant progress required in robust platforms for reliable and efficient chemometric testing of flexible and reconfigurable development systems that can be readily tailored towards different molecular analytes. This constitutes a key goal and formidable challenge in biosensor development. In this regard, we present Bio-Sensei, a fully automated High-Through-put platform that ‘guides’ the development of membrane biosensors to engineer chemically active surfaces. Its automated chemical handling, reconfigurable microfluidics & instrumentation powered by AI revolutionizes the biosensor development process. The platform integrates a robotic sampler arm, electrochemical, and digitized fluorescence setup, functioning as an Internet of Things (IoT) platform with a data analysis pipeline for "intelligent" development.

The instrumentation is focused on the in-house designed novel microfluidic core that simultaneously harnesses electrochemical and fluorescence techniques to form an electro-fluorochemical cell that enables precise readings from femto-to-micro scales. Bio-Sensei conducts stability, sensitivity, and selectivity experiments on samples to generate large datasets without compromising on the accuracy, quality, and precision. It not only speeds up the practical workflow but also unlocks experiments that are simply beyond human intervention. Bio-Sensei is an end-to-end automated platform integrating measurement and chemical handling hardware with a data processing software layer, facilitating sophisticated assay development, configurable flow, and measurement sequences for faster insights in high-throughput screening experiments. One of the most novel features of the platform is that it allows deployment of trained AI models on edge devices, this is beneficial for real-time processing of external verification of clinical or real-world samples with low latency that is independent of a central network making it directly applicable to POC devices.

We demonstrate its utility by functionalizing a nanopore membrane with mATCUN peptide that selectively binds to Cu2+ ions. Unsupervised clustering techniques are integrated to identify 0.03 μm as the optimized pore size of the functionalized nanopore membrane. The biosensor's stability, selectivity, and sensitivity were evaluated, achieving Limit-of-Detection (LOD) values of 10-6 and 10-15 M with fluorescence and I-V measurements, respectively. Further, an assay is developed to compare the performance of 6-Carboxyfluorescein and Rhodamine B as a fluorescent tag on mATCUN in human serum samples. The acquired data is used to develop a deep learning model that is deployed on edge to predict Cu2+ concentrations that forms a verification layer to our experiments. The results show over 90% prediction accuracy and AUC values of over 0.9 for ROC (Receiver Operating Characteristic)on samples.