Physically motivated feature engineering for classification of fluorescent DNA-stabilized silver clusters

DNA-stabilized silver clusters (Ag_N-DNAs) are fluorescent nanomaterials whose optical properties are encoded by DNA sequence. Due to the nucleobase-specific interactions of silver atoms with DNA [1], DNA sequence can tune the cluster sizes of Ag_N-DNAs from about 10 to 20 atoms per cluster, corresponding to a wide range of fluorescence colors spanning the visible to near infrared spectrum. Recently, machine learning approaches have been used to learn how DNA sequence relates to Ag_N-DNA fluorescence emission wavelength, allowing predictive design of new Ag_N-DNAs with bright fluorescence in desired wavelength ranges [2]. In order to improve the classification accuracy of these methods, we have developed physically motivated features based on the first crystallographic structures reported for Ag_N-DNAs. By better capturing the occurrences of certain non-adjacent nucleobase motifs, the accuracy of assigning a color class to an input DNA sequence is increased by up to 18%. This work contributes to the development of biomolecules as templates for ultrafine metallic nanostructures.

[1] S. M. Swasey, L. E. Leal, O. Lopez-Acevedo, J. Pavlovich, and E. G. Gwinn, Sci. Rep. 5, 10163 (2015).
[2] S. M. Copp, S. M. Swasey, A. Gorovits, P. Bogdanov, and E. G. Gwinn, Chem. Mater. 32, 430 (2020).