Simultaneous clustering and estimation of additive shape invariant models for recurrent event data

Area of Science:

• Computational Neuroscience
• Statistical Analysis of Neural Data
• Systems Neuroscience

Background:

• Advanced technologies allow recording from large neuron ensembles, posing statistical analysis challenges.
• Neural recordings often involve randomized interventions and complex response patterns with varying latencies.
• Identifying specific neuronal responses to stimuli in superimposed data is difficult.

Purpose of the Study:

• To develop a statistical model for analyzing neural spiking activity from large ensembles.
• To identify neuronal groups with unique responses to randomized stimuli and estimate these responses.
• To address challenges posed by superimposed neural responses and varying response latencies.

Main Methods:

• Introduction of a novel additive shape invariant model.
• The model simultaneously handles multiple clusters, additive components, and unknown time-shifts.
• Identifiability conditions for model parameters were established; algorithm properties examined via simulations.

Main Results:

• The proposed method was applied to neural spike train data from mice.
• Analysis of data from a visual discrimination task revealed three distinct functional neuron groups.
• These groups showed heterogeneous response patterns, including stimulus-specific activity and timing variability.

Conclusions:

• The novel additive shape invariant model effectively analyzes complex neural ensemble activity.
• The method successfully identifies neuronal subpopulations and their stimulus-evoked response characteristics.
• Findings provide insights into neural processing during sensory tasks and guide future experimental design.