Compact CLIP Model: Predicting Spectral Properties of AgNCs Using DNA Template

Abstract

DNA-templated silver nanoclusters (AgNCs) are a promising class of fluorescent nanomaterials synthesized using DNA as a template . These nanomaterials exhibit high quantum, excellent biocompatibility, and tunable fluorescence properties. Their structural stability and cost-effectiveness make them ideal candidates for applications in cell imaging and molecular sensing. The inherent variability and complexity of DNA sequences pose significant challenges to accurately predicting their corresponding fluorescence spectral properties. This study addresses this challenge by developing a deep learning-based compact Contrastive Language-Image Pre-Training (CLIP) model and constructing a comprehensive AgNCs database comprising 3844 samples. This model enables the prediction of fluorescence properties for AgNCs with diverse DNA sequences and structures. The model’s prediction accuracy was experimentally validated using 32 test samples, achieving a 44.85% accuracy rate, which is 15.9 times higher than random chance. The compact CLIP model effectively identifies AgNCs’ spectral emission peaks and intensities, with peak positions and fluorescence accuracy reaching 68% and 80%, respectively. This demonstrates its potential to guide the synthesis and simulation of DNA templated AgNCs. This approach is expected accelerate the development of cost-effective and efficient molecular signal characterization systems, opening new avenues for research in biochemical sensing, molecular computing, and other related fields.