MicroRNAs (miRNAs) have emerged as potential biomarkers for human diseases. Classification models established on the expression levels of multiple miRNAs can be used to realize a reliable diagnosis of diseases. The detection of expression patterns of multiple miRNAs using traditional methods is generally cost-prohibitive, labor-intensive, and limited by the requirements of sophisticated facilities. DNA molecular computing technology attempts to implement a multiple-miRNA-based disease classification model through reactions among DNA probes and miRNAs without the need to quantify each individual miRNA, and hence has the potential to realize fast and cheap clinical diagnosis of early-stage diseases. However, existing DNA molecular computing frameworks for disease diagnosis struggle in faithfully implementing even linear classification models with nonzero classification thresholds, which hinders their practical clinical applications. Herein, for the first time, a novel DNA molecular computing platform enabling full and accurate implementation of multiple-miRNA-based disease classification models is proposed. It consists of "multiplication," "addition," "calibration," and "diagnosis" modules. The first two modules are implemented experimentally through "one-pot" reactions among DNA probes and miRNAs, while the last two modules are carried out in silico. Experimental results have demonstrated that such a hybrid implementation strategy breaks through the bottleneck of conventional DNA molecular computing frameworks for disease diagnosis, and faithfully implements a disease classifier for discriminating diseased samples from normal ones. The new platform has the advantages of reduced design complexity, easy experimental implementation, and robustness to variations in signal amplification factors across samples, and opens up an avenue for point-of-care clinical diagnosis of early-stage diseases.
Tang et al. (Tue,) studied this question.