Ordinal Discriminative Dimensionality Reduction of Functional Profiles

Optical biosensors, utilizing biological components like DNA and antibodies, are effective analytical tools. However, analyzing sensor signals, especially in large datasets such as reflectometric imaging sensors, is computationally demaning both in terms of time and of memory usage. In this communication, we employ data from a reflectometric imaging sensor to track the antibody-antigen reaction progression using video images of the biosensor surface. Analyzing temporal changes in light intensity provides insights into reaction progression; however, the need for an automatic detector of biological process reaction requires dimensionality reduction of the data from the sensor, which arrive in the form of functional profiles from video signals. We illustrate a workflow which includes cleaning light disturbances, condensing video data, and reducing the dimensionality of the obtained functional data. Departing from traditional methods like Functional Principal Component Analysis (FPCA), we discuss functional-ordinal Canonical Correlation Analysis to optimize correlation between the high-dimensional predictor and the outcome. The conclusion highlights that, the combination of preprocessing and CCA for effective discrimination among different reagent concentration levels, enables one to project the video signal into a 2-dimensional space. This innovative approach enhances our ability to detect virus vitality in biomanufacturing processes.