Nanopore sensing is a powerful technique for the detection and study of single molecules. The underlying resistive pulse principle relies on molecules modulating an ionic current as they pass through a nanoscale constriction between two reservoirs. The technique has recently been extended to the detection of modifications attached along a single molecule composed of a double strand of DNA. As the strand threads through the nanopore, attachments produce secondary peaks in the already reduced current. The ability to encode information in modification sequences and self-assemble designed strands through DNA nanotechnology opens up promising possibilities ranging from highly parallel single-molecule sensing to DNA data storage. This work investigates experimental methods and analysis techniques to improve the readout of information contained in the position and type of modifications attached along the strand.

Experimentally, it is found that fluid flows opposing the movement of DNA strongly promote single-file, unfolded entry into conical glass nanopores. The effect can be induced both with flows due to externally applied pressure and electro-osmotic flows (EOF) along the pore walls. The ability to prevent folded entry into the pore is crucial for decoding of attached modifications as it facilitates the identification and classification of secondary current peaks.

In the investigation of analysis methods, Bayesian inference is assessed as a tool to extract information from the current signal. While useful to detect a strand’s folding state, it is found that deep learning-based approaches exhibit superior accuracy where enough training data is available. Two convolutional neural network (CNN) architectures are presented, the first one targeted at the classification of previously known modification sequences and the second one for the identification of modifications at arbitrary positions.

The last chapter shows that the deep learning approach is capable of classifying translocation events in real time. A probability calculation reveals how many events are required for a measurement to conclusively show the presence of a certain molecule. Re-measuring molecules by driving them in and out of the nanopore is investigated as a method to improve modification decoding. The thesis concludes with a critical discussion of the presented techniques and their contribution to the development of sensing and data storage applications built on nanopore-based readout of structural modifications on DNA strands.