Machine learning for control over the colors of fluorescent silver nanoclusters

Engineered DNA sequence has shown potential to realize self-assembling DNA nanostructures with custom geometries that can direct assembly of molecules and nanoparticles, and even program motion. Further, with advances in nanotechnology, DNA has shown promising attributes that can enable it to be used to template inorganic particles, such as DNA-stabilized silver clusters that comprise of just ~10-30 atoms. So far, the aforementioned silver clusters are the most studied DNA-stabilized metal cluster as they exhibit narrow-band fluorescence colors spanning visible to near-infrared wavelengths. Consequently, they have an intriguing potential as fluorescence reporters for a wide range of nanoscale events. Unfortunately, their applications are inhibited by the unsolved puzzle of the cluster “genome”: How does DNA sequence determine silver cluster color?

Materials informatics is emerging as an efficient route to realizing desired material properties through data-driven prediction of synthesis parameters. Therefore, it is imperative that machine learning and bioinformatics be employed hand-in-hand to subdue the challenges enlisted and also uncover how DNA sequence selects for DNA-stabilized silver clusters.

Recently, a team of researchers including Dr. Stacy M. Copp from the Center for Integrated Nanotechnologies, Los Alamos National Laboratory, in collaboration with Alexander Gorovits ( PhD candidate), Sruthi Gudibandi and Dr. Petko Bogdanov at the University at Albany – SUNY and Dr. Steven M. Swasey and Professor Elisabeth Gwinn at University of California Santa Barbara carried out an investigation on the genomic role of DNA sequence for fluorescent silver clusters using a data-driven approach. Specifically, they sought to engineer DNA templates so as to control the morphology, and thereby the fluorescence color, of clusters of just about 10-20 silver atoms. Their work is currently published in the research journal, ACS Nano.

In brief, their research work entailed the application of high-throughput experimental methods to generate a large training data set that associated many DNA sequences with the fluorescence spectra of the silver clusters they stabilized. Further, they utilized a pattern recognition algorithm to extract DNA base subsequences termed ‘motifs’ that were correlated to specific fluorescence wavelength bands and employed in a machine learning classifier. Lastly, the researchers used the trained classifier to select new templates for the DNA-stabilized silver clusters in specific color bands, and also verify the templates experimentally.

The authors observed that the technique they employed first discovered the DNA base motifs that were most discriminative of color, using a combination of motif mining and feature selection. Moreover, the set of trained pairwise classifiers identified new candidate sequences built from the discovered motifs that fell within a selected color class. Furthermore, they realized that their technique improved selectivity of longer wavelength DNA-stabilized silver clusters, at the boundary of the visible and near-infrared spectrum.

In summary, Copp and colleagues successfully developed a data-driven technique to design DNA templates that select silver clusters of certain sizes, and therefore fluorescence colors, by learning from a large experimental training data set. In general, their technique demonstrated successful design of DNA templates that discriminated for cluster size differences of just a few silver atoms, as well as exploring why the discovered base motifs were selective of color. Altogether, their work portrayed the power of combining new advances in high-throughput experiments with data informatics to achieve molecular design of materials systems.

“We believe that these results have implications not just for DNA-templated metal clusters but also for other soft matter systems characterized by extremely large parameter spaces, where techniques such as machine learning are beginning to transform the way we study such systems.” Said Stacy Copp to Advances in Engineering.