A novel mathematical approach to the study of behavioral readout

Precise quantification of how spatio-temporal structures in neural activity are consequential for behavior, remains challenging. In this work I introduced a novel mathematical method to investigate behavioral readout. The method is based on the embedding of neural activity patterns into a metric space and on the definition of perceptual distances, that describe the internal representation of percepts. While being rigorously and mathematically well defined, the method is very general and flexible in the definition of perceptual distances. This makes it ideally suitable to the study of various systems, starting from single cells to neural populations. As a proof of principle, I applied the presented method to experimental data provided by Edmund Chong and Dmitry Rinberg at the Neuroscience Institute at New York University (NYU). We combined the novel method with an interventional approach to investigate the neural code used in the mice olfactory bulb (OB) to generate behavior. We trained animals to recognize artificial stimulation patterns on the OB and we systematically varied the stimulus features in order to thoroughly investigate the neural activity space. We were able to show some basic elements of OB coding: both spatial and temporal features of OB activity patterns are read out by animals, the readout favors early activity (primacy effect), spatial features are combined in an almost linear fashion, while temporal features are used in a more complex way. Our findings were replicated using logistic regression, a classical method, but our approach was advantageous in terms of model complexity and interpretability.