Identification and Classification of Functional Split G-Quadruplexes Using Machine Learning-Guided Activity Screening

Abstract

Split G-quadruplexes are considered excellent tools for biosensing and diagnostics, but splitting G-quadruplexes may often lead to a loss of function, limiting their effectiveness. This study aims to identify and classify functional split G-quadruplexes based on the ability of the G-quadruplex motif to generate a fluorescence turn-on response and undergo phase separation. A series of split G-quadruplexes were designed, and their characterization was conducted using fluorescence spectroscopy, fluorescence microscopy, UV-vis spectroscopy, and circular dichroism to investigate their functional properties (fluorogenic response, phase separation, and DNAzyme activity). Multivariate analysis and machine learning-based pattern recognition revealed that structural changes due to the splitting of G4-forming sequences correlate with their ability to form phase-separated condensates, which enhance their fluorogenic and DNAzyme activity. The machine learning-based activity screening was used to identify split G-quadruplexes, which may have high, moderate, or low functional activity. This integrative approach provides a predictive framework for engineering functionally active split G-quadruplexes and establishes a platform for their application in molecular diagnostics.

Keywords: DNAzyme; machine learning; molecular diagnostics; phase separation; split G-quadruplex.