Anomaly detection in sequential single-molecule data

Single-molecule measurements, for example current-time or current-distance recordings from single-molecule detectors, usually produce data that are characterised by large variance and high levels of "noise". These characteristics are to some extent intrinsic to the fundamental physical properties of the system under study and they can contain important information about the structure and dynamics of a molecular system. So extracting the full information content of the data, preferably in an unsupervised manner without a priori assumptions, is really critical. Accordingly, recent years have seen the emergence of advanced data analysis tools in this field, ranging from more conventional dimensionality reduction techniques to Deep Learning and image recognition.

This project will build on these advances and explore an entirely new concept: rather than asking what a molecular event might look like, we will ask "what is not background?" (and by implication define what events or "anomalies" in the data are). We will combine statistical methods and our understanding of single-molecule data with different neural network architectures, such as recurrent neural networks (RNN) and generative adversarial networks (GAN), and develop a toolkit for next-generation single-molecule data analysis. We are closely aligned with Birmingham's Interdisciplinary Data Science Institute, the Schools of Mathematics and Computer Science and the Turing Institute, which offer great opportunities for further collaboration. So if you have a background in Chemistry, Physics, Mathematics, Computer Science or a related area, are enthusiastic about working at the interface of Computer Science and Nanoscience and in an interdisciplinary team, then please get in touch!