Machine learning-based fluorescence sensor array: Accurate discrimination, quantitative assay of phenothiazine drugs via versatile DNA probes

Abstract

Phenothiazine drugs are widely used to treat psychotic diseases in humans. The phenothiazine ligand enables unique interaction between the drugs and DNA double helix. We discovered that it was impossible to select aptamer to selectively bind with phenothiazine drug. In this study, an alternative fluorescence sensor array was first developed to accurately discriminate four typical phenothiazine drugs based on machine learning. As the eight sensor elements, versatile DNA probes were designed based on different oligonucleotides, two unique fluorescence dyes, and diverse buffer solutions. The fingerprint data can offer the fluorescence response patterns against each phenothiazine drug. Linear discriminant analysis (LDA) can discriminate single one or mixture samples of phenothiazine drugs with 100% accuracy. Besides, one neural network model showed a good linear relationship between predicted and actual concentrations to quantitatively assay specific phenothiazine drugs. The proposed fluorescence sensor array presents a new way to accurately discriminate phenothiazine drugs in pharmacy or biology field.