Probabilistic Graphical Models of Functional and Effective Neuronal Connectivity

• Activity Number: 114
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Seif Eldawlatly and Mehdi Aghagolzadeh and Karim Oweiss*+
• Companies: Michigan State University and Michigan State University and Michigan State University
• Keywords: graphical models ; neuronal connectivity ; spike trains

Abstract

Coordination among cortical neurons plays an important role in mediating perception, cognition and motor actions. Deciphering the underlying neural circuitry is therefore of utmost importance in basic and clinical neuroscience. Probabilistic graphical models are powerful tools that can infer statistical relationships between the spiking patterns of simultaneously observed neurons. We present two methods for analyzing these patterns in multiple sensory systems based on graphical models. The first method employs dynamic Bayesian networks (DBNs) for inferring the effective connectivity among neurons. Unlike traditional pairwise correlation metrics, DBN accounts for precisely timed spiking of the entire neuronal population, allowing it to identify direct, possibly nonlinear, coupling between neurons by explaining away unlikely causes of firing. The second method identifies the functional connectivity between the neurons using a minimum entropy distance (MinED) method. MinED extends the well known maximum entropy models to model higher order interactions among neurons using hypergraphs. We demonstrate the use of both methods using data recorded from somatosensory and visual cortices.