Reading-out DNA translocation experiments with (un)supervised Machine Learning

DNA molecules can electrophoretically be driven through a nanoscale opening in a material giving rise to rich and measurable ionic current blockades. In this work, we train machine learning (ML) models on experimental ionic blockade data from DNA nucleotide translocation through 2D pores of different diameters. We propose a novel method that at the same time reduces the current traces to a few physical descriptors and trains low-complexity models. We describe each translocation event by four features referring to the structure of the ionic current time series. Training on this lower dimensional data and utilizing deep neural networks (DNN) and convolutional neural networks (CNN) we reach a high accuracy. Compared to more complex baseline schemes such as extreme gradient tree boosting (XGBoost) and recurrent neural network (RNN) based models trained on the full ionic current trace the former perform is comparable or even better. Our findings clearly reveal that the use of the ionic blockade height as a feature together with a proper combination of neural networks and feature extraction provides a strong enhancement in the detection and read-out sensitivity of novel nanopore sequencers.