There are a vast number of pathogens that impact global public health, necessitating an accessible assay capable of detecting multiple targets simultaneously. Lateral flow assays (LFAs) have the potential to fill this role as a cost-effective, rapid, and simple technology instrumental in the detection of many analytes. However, multiplexed detection using nucleic acid LFAs is difficult due to the increased chance of non-specific binding as more targets are added to the assay. In this work, we aim to increase specificity and multiplexing potential in LFAs. We first showcase the process of developing a nucleic acid LFA by evaluating both fluorophores and gold nanoparticles to generate a visible signal. As fluorophores require a fluorescent light source, we moved forward with gold nanoparticles, which have a readout visible to the naked eye. Additionally, we automated the LFA fabrication process using an Echo Liquid Handler. Finally, we assessed methods to convert double-stranded to single-stranded DNA, required for compatibility with LFAs. In the future, we look to optimize signal visibility while increasing multiplexability. This work highlights the potential of multiplexed LFAs as a robust technology capable of significantly improving public health responses and outcomes.

The 4-letter genetic alphabet found in Nature is the fundamental basis of biological information storage. As synthetic biologists continue to manipulate the genetic alphabet, they have begun to push against the boundaries of nature itself. Unnatural base-pairing xenonucleic acids (XNAs) are synthetic nucleotides that can pair orthogonally with the standard bases. By increasing chemical and structural diversity, XNAs are poised to enable a plethora of next-generation biotechnologies, including XNA-containing nucleic acid therapeutics (XNAptamers), catalytic nucleic acids (XNAzymes), and an expanded genetic code through a larger codon table. Although the potential of XNAs is near-limitless, the infrastructure required to study XNAs, notably sequencing, is antiquated. Previously, the Marchand Group leveraged commercial nanopore sequencing devices from Oxford Nanopore Technologies to sequence XNAs. This process outputs characteristic current signals that need to be decoded or “basecalled.” The first XNA basecallers used statistical k-mer models to decode XNA containing current signals, yet, their basecalling accuracy is a far cry from commercial basecallers (k-mer: 60-80%, commercial: >95%). Modeling our approach after commercial DNA basecallers, we have built a binary classification training pipeline that leverages long short-term memory (LSTM) neural networks and commercial nanopore sequencing to achieve more precise sequencing of XNAs. Thus far,  we have built models to effectively basecall three XNA base pairs with varying motivations: B≡Sⁿ for studying XNA replication fidelity in PCR due to high error rates, and P≡Z/Ds:Px for their unnatural functional groups (e.g. nitro groups and hydrophobicity) making them useful for applications such as XNAptamers. Currently, our binary classification models have testing accuracies as high as around 95% and we look to further improve our training methods through new model architectures such as transformers. Moving forward, we look to expand our basecaller to perform multi classification, allowing for generalized, de novo basecalling similar to commercial basecallers.