Applying Machine\nLearning and SERS for Precise Typing\nof DNA Secondary Structures

Surface-enhanced Raman spectroscopy (SERS) has been demonstrated\nas an effective method for elucidating secondary structural characteristics\nof DNA. However, the inherent complexity of the DNA conformation and\nthe lack of SERS samples pose challenges for identifying numerous\nsecondary structures. To address these issues, a synergistic method\nintegrating machine learning with SERS was proposed so as to analyze\nthe SERS spectra of 54 well-defined conformational oligonucleotides,\nnamely, G-quadruplex (G4), i-motif (iM), double-strand (DS), and single-strand\n(SS) configurations. Principal component analysis (PCA) effectively\nsegregated the oligonucleotides into three distinct conformational\ngroups (G4s, iMs, and others). Furthermore, linear discriminant analysis\n(LDA), K-nearest neighbor (KNN), and support vector machine (SVM)\napproaches were utilized to improve the typing accuracy of 54 trained\nsequences. This enabled the correct classification of the structures\nof five untrained sequences, as well as the identification of the\npredominant conformations including G4, iM, and DS formed by two complementary\nG-rich and C-rich sequences in acidic and neutral pH conditions. The\nresults of this study demonstrated the potential of the proposed methodology\nfor rapid screening and prediction of secondary DNA conformations.