Guest Presenters: Dr. Ruben Sanchez-Romero (Post-doc working with Dr. Michael Cole at Rutgers University)
& Kirsten Peterson (Graduate student working with Dr. Michael Cole at Rutgers University)

Dr. Ruben Sanchez-Romero

Talk Title: Activity flow modeling approach of Cole Lab.

Kirsten Peterson

Talk Title: Estimating functional connectivity (FC) using regularized partial correlation or multiple regression

Abstract: I will be presenting on how to estimate functional connectivity (FC) using regularized partial correlation or multiple regression. Partial correlation and multiple regression FC are advantageous over the field-standard method of pairwise Pearson correlation in that they estimate direct connectivity, free of confounded and indirect connections. However, partial correlation and multiple regression FC models are susceptible to overfitting to noise, which can render their estimated FC weights unstable and noisy (i.e., unreliable). Adding regularization can reduce this overfitting to yield FC estimates that are both causally valid and reliable. I will especially discuss lasso/graphical lasso (L1-regularized regression/partial correlation), which have been the most successful in our validation tests. While we have focused on applying these methods to fMRI data, we expect that other modalities can benefit from them as well.