Unsupervised data analysis for virus detection with a surface plasmon resonance sensor

We propose an unsupervised approach for virus detection with a biosensor based on surface plasmon resonance. A column-based non-negative matrix factorisation (NNCX) serves to select virus candidate time series from the spatio-temporal data. The candidates are then separated into true virus adhesions and false positive NNCX responses by fitting a constrained virus model function. In the evaluation on ground truth data, our unsupervised approach compares favourably to a previously published supervised approach that requires more parameters.