Artificial Intelligence-Assisted Label-Free Spectroscopic Quantification of Global DNA Cytosine Methylation in a Miniature Plasmonic Pickering Emulsion

Epigenetic DNA methylations are early and frequently observed events in a diversity of diseases such as cancer. Despite the considerable clinical values for cancer liquid biopsy, quantitative analysis of DNA methylations remains a major challenge due to the lack of rapid, sensitive detection techniques. Here, an artificial intelligence-assisted label-free surface-enhanced Raman spectroscopy (SERS) (iMeSERS) biosensor is reported for simultaneous quantification of C 5 -methylcytosine ( 5m C) level and methylation ratio in DNA samples. This method utilizes the plasmonic Pickering emulsions as the biosensing platform for label-free SERS detection, formed upon the addition of a sub-microliter DNA sample to the hydrophobic Au nanostar-containing n -decane. Distinct spectral signatures of monophosphates of canonical deoxyribonucleotides (dNMPs) and the common methylation modification 5-methyl-2′-deoxycytidine (d 5m CMP) are identified and distinguished by the iMeSERS biosensor. The deep learning algorithms trained with SERS signatures of dNMPs and d 5m CMP are then applied to the quantitative analysis of global DNA methylation. The exceptional capability of the deep learning-driven approach is demonstrated for simultaneous quantification of the methylation ratio and level using a sub-microliter volume of DNA samples. This work shows the power of label-free SERS techniques combined with deep learning algorithms for quantitative analysis of epigenetic DNA modifications with great promises for clinical diagnosis.