Identification of amino acids with sensitive nanoporous MoS2: towards machine learning-based prediction

Protein detection plays a key role in determining the single point mutations which can cause a variety of diseases. Nanopore sequencing provides a label-free, single base, fast and long reading platform, which makes it amenable for personalized medicine. A challenge facing nanopore technology is the noise in ionic current. Here, we show that a nanoporous single-layer molybdenum disulfide (MoS2) can detect individual amino acids in a polypeptide chain (16 units) with a high accuracy and distinguishability. Using extensive molecular dynamics simulations (with a total aggregate simulation time of 66 µs) and machine learning techniques, we featurize and cluster the ionic current and residence time of the 20 amino acids and identify the fingerprints of the signals. Using logistic regression, nearest neighbor, and random forest classifiers, the sensor reading is predicted with an accuracy of 72.45, 94.55, and 99.6%, respectively. In addition, using advanced ML classification techniques, we are able to theoretically predict over 2.8 million hypothetical sensor readings’ amino acid types. Molecular dynamics simulations combined with machine learning techniques enable the prediction of MoS2 nanopore sequencing capabilities. A team led by N. R. Aluru at the University of Illinois at Urbana-Champaign used logistic regression, nearest neighbor, and random forest classifiers to develop a machine learning-based platform capable of predicting the sensing capabilities of nanoporous, atomically thin MoS2. The material was shown to be able to identify individual amino acids in polypeptide chains with high accuracy and distinguishability. Twenty amino acids could be detected and categorized in different classes based on current-residence time training data, with an accuracy of up to 99.6%. These results show promise for the development of amino acid detection platforms with atomically thin materials assisted by machine learning.