DRIADA: BRIDGING SINGLE-NEURON SELECTIVITY AND POPULATION GEOMETRY IN BEHAVING MICE

Understanding how neural populations represent information requires bridging two traditionally separate scales of analysis: individual neuron selectivity and collective population geometry. While single-neuron approaches identify which cells encode specific variables, and population-level methods reveal low-dimensional manifolds, the connection between these scales remains poorly understood. We present DRIADA (Dimensionality Reduction for Integrated Activity Data), a unified computational framework that explicitly links single-cell selectivity to population-level representations in behaving mice. DRIADA combines single-neuron and population analyses in an integrated workflow. The INTENSE module identifies which neurons encode specific behavioral variables by measuring how much information neural activity carries about behavior. This information-theoretic approach detects both simple and complex relationships that traditional correlation methods miss. A two-stage statistical procedure ensures robust detection while controlling for multiple comparisons and accounting for temporal delays between neural activity and behavior. Critically, INTENSE includes a disentanglement stage that separates true selectivity from spurious correlations when neurons respond to multiple correlated behavioral variables simultaneously. At the population level, DRIADA characterizes the geometry of neural activity spaces and extracts latent behavioral variables from collective dynamics. The framework then connects these scales by revealing how individual neurons contribute to population-level representations. We demonstrate DRIADA capabilities using hippocampal calcium imaging data from freely behaving mice, showing how individual neuronal selectivity relates to the structure of population activity manifolds. The work was supported by Non-Commercial Foundation for Support of Science and Education "INTELLECT".