Nonlinear Classifiers Based on DNA Logic Circuits for Cancer Diagnosis

Abstract

DNA logical circuits can be applied to accurate classification of cancer status, benefiting from their excellent biocompatibility and parallelism. However, the existing cancer diagnosis models based on DNA logic circuits mainly adopt a linear structure, which makes it difficult to fully capture the complex nonlinear distribution characteristics in the disease data. In addition, DNA logic circuits cannot directly sense the expression levels of microRNAs (miRNAs). Here, we constructed a nonlinear classifier based on DNA logic circuits with the random forest algorithm. The classifier can directly sense the expression level of miRNAs in serum samples without isolating specific miRNAs and transmit the signals to the logic classification module and complete the nonlinear classification of cancer status. We validated the classification performance of the constructed nonlinear classifiers by using miRNA expression level samples to diagnose adenocarcinoma, ductal and lobular neoplasms, and squamous cell carcinoma with accuracies of 95.4%, 96.6%, and 97.2%, respectively. The classification results generated using the nonlinear classifiers based on DNA logic circuits showed a strong agreement with the actual disease states labeled in TCGA, as well as with the random forest algorithm, and had high parallelism and stability in the multiclassification of three different cancers. This work shows the great potential of DNA logic circuit-based nonlinear classifiers in cancer diagnosis, which provides a new approach to design efficient, accurate, and intelligent integrated disease diagnosis schemes.