Summary
Brains are not engineered solutions to a well-defined problem but arose through selective pressure acting on random variation. It is therefore unclear how well a model chosen by an experimenter can relate neural activity to experimental conditions. Here we developed “Model-free identification of neural encoding (MINE)” using convolutional neural networks (CNN) to relate aspects of tasks to neural activity. Although flexible, CNNs are difficult to interpret. We use Taylor decomposition approaches and explainable machine learning techniques to understand the mapping from task features to activity. We apply MINE to a published cortical dataset as well as experiments designed to probe thermoregulatory circuits in zebrafish. MINE allowed us to characterize neurons according to their receptive field and computational complexity, features which anatomically segregated in the brain. We also identified a new class of neurons that integrate thermosensory and behavioral information which eluded us previously when using traditional clustering and regression-based approaches.
Competing Interest Statement
JDC is employed by Hitachi Solutions, Irvine CA.
Footnotes
A comparison of MINE and spike triggered analysis has been performed. This led to addition of a new figure 4, new figure S4, a new results section, a new discussion paragraph and a new methods section.