# Machine Learning-Assisted DNA Origami Shape Sorting Using Fingerprinting Nanosensors and Feature Engineering

## TL;DR

This paper introduces a machine learning method combined with nanosensors to accurately identify different DNA origami shapes, offering a low-cost alternative to traditional imaging techniques.

## Contribution

The novel integration of a nanosensor array with machine learning for DNA origami shape sorting is presented.

## Key findings

- A nanosensor array with 11 fluorophore-labeled DNA probes was used to detect DNA origami shapes.
- Machine learning achieved 94% accuracy in distinguishing DNA origami triangles, nanotubes, and scaffold strands.
- The method enables high-throughput, label-free classification of DNA nanostructures.

## Abstract

Reconfigurable DNA nanostructures have emerged as a promising
research
area with applications in drug delivery, molecular computing, biosensing,
and stimuli-responsive soft nanomaterials. While significant progress
has been made in creating novel DNA nanostructures and exploring their
applications, comparatively little effort has focused on developing
new methodologies to confirm their folding. Conventional imaging approaches
typically rely on sophisticated microscopy techniques including atomic
force and transmission electron microscopy. Alternative low-cost methods
for verifying DNA nanostructure assembly and shape sorting are thus
highly valuable. Here, we present a fingerprinting nanosensor array
integrated with machine learning (ML) to distinguish between two DNA
origami shapes (triangle and nanotube) and to differentiate them from
an unfolded scaffold strand. The nanosensor array, consisting of 11
nanoassemblies, termed nanosensors, is prepared by complexing graphene
oxide nanosheets (nGO) with 11 fluorophore-labeled single-stranded
DNA probes. Upon complexation, the fluorescence of the DNA probes
is quenched through graphene oxide-mediated fluorescence quenching.
Adding the DNA nanostructures to each nanosensor displaced a fraction
of the surface-adsorbed fluorescent DNA probes, producing unique fluorescence
recovery signatures that were subsequently processed through feature
engineering for accurate ML-assisted classification. Using this approach,
we achieved 94% prediction accuracy in discriminating DNA origami
triangle, DNA origami nanotube, and unassembled M13 scaffold. Our
strategy provides a new and generalizable platform for shape sorting
in DNA origami field, offering new avenues for high-throughput, label-free
classification.