Files
Abstract
Rapid and sensitive virus detection is important to prevent, control, and manage epidemics or pandemics. This dissertation focuses on exploring several rapid detection strategies by combining surface-enhanced Raman spectroscopy (SERS) and advanced machine learning algorithms (MLAs) for SARS-CoV-2 diagnostics and demonstrates its capability of classification and quantification of different respiratory viruses and virus mixtures.Three strategies have been proposed to detect SARS-CoV-2 from clinical human nasopharyngeal swab (HNS) specimens. The first strategy is to use DNA probes modified silver nanorod array substrates to capture the RNAs of SARS-CoV-2. SERS spectra have been collected after RNA hybridization, and a recurrent neural network (RNN)-based deep learning model is developed to classify positive and negative specimens, with an overall classification accuracy of 98.9% and 98.6% in blind testing. The second strategy is to use an ACE2 capture protein functionalized SERS sensor to specifically capture the intact virus and its variants. A convolutional neural network (CNN) model has been developed to simultaneously classify and quantify the coronavirus variants, achieving an accuracy of > 99%. Finally, a direct diagnostic strategy is designed to detect inactivated HNS specimens by combining SERS and SFNet model, which can achieve an overall 98.5% accuracy in distinguishing positive or negative specimens and predict the cycle threshold (Ct) accurately based on reverse transcription-polymerase chain reaction (RT-PCR). All the detections can be accomplished within 25 min.
The SERS+MLA strategy can also be extended to detect a panel of 13 respiratory virus species. Various MLAs have been employed to differentiate the viruses based on their SERS spectra, achieving an accuracy exceeding 99%. A two-step classification and quantification strategy has been demonstrated to effectively predict virus concentrations in buffer and saliva. Finally, a similar detection and spectral analysis strategy has been demonstrated and can be effectively used to identify and quantify viruses in a mixture, yielding results aligned well with the experimental design.
All the results obtained in this dissertation demonstrate that combining SERS and MLA could serve as a rapid and potential diagnostic platform for point-of-care, and it may have the advantage of complimenting the PCR method.