Machine learning-assisted detection of single-point mutations via DNA-templated gold nanoparticle growth

Abstract

Detecting single point mutations, such as PIK3CA mutations, is vital for precision diagnostics but remains challenging due to subtle sequence differences. This study introduces a machine learning-assisted colorimetric biosensor that utilizes DNA-templated gold nanoparticle (AuNP) growth for sensitive and specific single point mutation detection. The biosensor employs a hairpin DNA probe that selectively hybridizes with mutant PIK3CA sequences. In its native state, the loop region catalyzes the reduction of Au ions, forming a stable, red-colored AuNP solution. Hybridization with the mutant sequence disrupts the loop, reduces nucleobase accessibility, and induces nanoparticle aggregation, resulting in a visible color shift to purple. Molecular dynamics (MD) simulations show that 4.5 nm AuNPs preferentially bind within the intact hairpin loop due to size complementarity, while smaller nanoparticles interact within DNA grooves. Mutation-induced disruption alters this templating effect, leading to heterogeneous nanoparticle growth. The key parameters, including DNA probe and gold ion concentrations, pH, and temperature, were optimized using Plackett–Burman and Box–Behnken designs. The optimized biosensor demonstrated a detection range of 10–1000 nM and a detection limit of 8.13 nM for PIK3CA mutations. To enable portable, real-time analysis, the biosensor was integrated with a smartphone-based machine learning platform. RGB images of AuNP solutions were analyzed using supervised learning, with the random forest regression model providing high recovery rates and low prediction errors. By integrating molecular modeling, experimental optimization, and AI-driven analysis, this work offers a scalable and cost-effective single point mutation detection platform with strong potential for personalized diagnostics and point-of-care testing in resource-limited settings.