Abstract
Abstract
DNA-based computing enables the processing of complex molecular information within a biological context, particularly suitable for diagnostic and therapeutic applications. However, the complexity of molecular dynamics of DNA-based chemical reactions poses challenges to integrating large-scale neural networks with complex topologies, which restricts their effective feature recognition. Here, we present a hybrid bioelectronic computing architecture, the DNA-encoded microfluidic integrated computing (DMIC) chip to enhance the scalability and complexity of DNA neural networks for in-molecule machine learning. The DMIC chip employs controllable droplets as micro-containers to store, transport, and process DNA elements, coupled with a 2D electrode array that precisely organizes DNA-encoded droplets array. This design enables the formation of complex DNA reaction networks and supports efficient on-chip computation through targeted droplet movement and reconfigurable fluidic pathways. Using the DMIC chip, we achieved the monolithic integration of up to twenty-six DNA neurons, which learn and model patterns of simple to moderate complexity. Programmable droplet pathways support the execution of machine learning algorithms, including support vector machines (SVM), convolutional neural networks (CNN), and fully connected neural networks (FCNN). The DMIC chip supports in-situ learning via on-chip DNA convolutional neural networks trained, achieving the classification of three 12-bit binary patterns. Furthermore, we demonstrate the implementation of the DMIC chip to simultaneously analyze thirty-six disease-related microRNAs for multi-class diagnosis of three cancer types. This integrated computing architecture offers exciting potential for large-scale molecular data processing and diagnostic applications.