Rarely categorical, highly separable representations along the cortical hierarchy

Key Points:

• The study utilized the International Brain Laboratory (IBL) public dataset, including task, behavioral, and electrophysiological recordings segmented into trials, focusing on variables like block prior, stimulus contrast/location, choice, and movement metrics.
• Sessions and trials were selected based on completeness of behavioral and electrophysiological data, excluding trials with delayed responses or from balanced blocks to avoid artifacts; neurons were included if their firing rates were within specified thresholds and their activity was well-predicted by a reduced-rank regression (RRR) model.
• The RRR encoding model was developed to analyze time-varying neural responses as linear combinations of task and movement variables, employing shared temporal basis vectors to reduce parameters and improve interpretability; model fitting used ridge regression with cross-validation to optimize hyperparameters.
• Selectivity profiles for individual neurons were computed from regression coefficients, enabling clustering analyses in selectivity or condition spaces to identify functional groups, with statistical controls to avoid biases from session-specific clusters or low-selectivity neurons.
• Population-level analyses included dimensionality estimation via participation ratio, cross-validated decoding of cognitive, sensory, and movement variables, and an iterative algorithm to identify independent conditions, supported by synthetic data simulations to explore the relationship between selectivity structure, dimensionality, and separability.