2024 AIChE Annual Meeting
Machine-Learning the Near-Infrared Fluorescence of DNA-Stabilized Silver Nanoclusters with Quantified Uncertainty and Explainability
Building on previous work by the Copp group at UC Irvine, we train and test an ensemble of gapped k-mer SVR’s (support vector regressors) to predict, with quantified uncertainty and explainability, the NIR fluorescence intensity of an AgN-DNA from its DNA sequence. Our training data constitutes ~2,400 examples of AgN-DNAs and their measured NIR emission intensities. The gapped k-mer kernel scores the similarity between a pair of DNA sequences according to matchings of gapped k-mers between them; a length-k DNA subsequence with wildcard positions. We expect our machine learning model to be useful for computationally-designing AgN-DNAs for bioimaging, circumventing a time- and cost-intensive trial-and-error search in the lab for these promising bioimaging agents.