Detecting effective connectivity in networks of coupled neuronal oscillators

Abstract

The application of data-driven time series analysis techniques such as Granger causality, partial directed coherence and phase dynamics modeling to estimate effective connectivity in brain networks has recently gained significant prominence in the neuroscience community. While these techniques have been useful in determining causal interactions among different regions of brain networks, a thorough analysis of the comparative accuracy and robustness of these methods in identifying patterns of effective connectivity among brain networks is still lacking. In this paper, we systematically address this issue within the context of simple networks of coupled spiking neurons. Specifically, we develop a method to assess the ability of various effective connectivity measures to accurately determine the true effective connectivity of a given neuronal network. Our method is based on decision tree classifiers which are trained using several time series features that can be observed solely from experimentally recorded data. We show that the classifiers constructed in this work provide a general framework for determining whether a particular effective connectivity measure is likely to produce incorrect results when applied to a dataset.

Conflict and adaptation signals in the anterior cingulate cortex and ventral tegmental area

“…However, the directionality of that communication was unclear. Therefore, to assess the directionality of communication between the ACC and VTA during these putative conflict and adaptation epochs, we implemented Boykin et al .’s 26 method of partial directed coherence (PDC; see Methods). The mean PDC in the 3–5 Hz frequency band (hence 4 Hz) was calculated for each maze region, trial, and animal, separately.…”

“…Our implementation of PDC followed Boykin, Khargonekar, Carney, Ogle, & Talathi’s 26 method. Briefly, we linearly detrended the unfiltered LFPs and then fitted relevant sections of the ACC and VTA LFPs to an autoregressive model where the maximum model order was determined according to Akaike’s 63 information criteria.…”

“…Every 12 trials (approximately), the rule switched and the rats needed to shift and sustain responses to the alternative side to obtain reward. We examined task-dependent changes in LFP power and coherence and assessed signal directionality between the ACC and VTA through the use of partial directed coherence (PDC) modelling 26 , 27 . On the basis of prior studies indicating that the ACC and VTA communicate via slow (~4 Hz) oscillations 28 – 30 and that such oscillations in the cortex are modulated by DA 18 , 31 , we hypothesised that 4 Hz VTA-to-ACC PDC model magnitudes would be maximal as an animal responded to error signals from the environment and subsequently adapted its behaviour.…”

Communication between the Anterior Cingulate Cortex and Ventral Tegmental Area during a Cost-Benefit Reversal Task

“…To further clarify the directionality of the VTA-ACC signaling as inferred from LFP data, we used PDC modeling (Baccalá and Sa-meshima, 2001;Boykin et al, 2012), which exploits the predictability of one brain area at one point in time by another brain area at a different time point in the frequency domain. The advantage of PDC modeling over other methods we have used to assess directionality, such as amplitude cross-correlations (Adhikari et al, 2010;Elston and Bilkey, 2017;Elston et al, 2018), is that it allows for testing the possibility of simultaneous, bi-directional communication, which is likely to occur between two reciprocally connected structures like the ACC and VTA.…”

“…Task-related, causal relationships between the ACC and VTA LFPs were assessed with a partial directed coherence (PDC) algorithm, a frequency-resolved estimate of Granger causality, which uses multivariate autoregressive modeling to exploit the predictability of information in one brain area by past activity in another (Baccalá and Sameshima, 2001). Our implementation of PDC followed Boykin, Khargonekar, Carney, Ogle, & Talathi's (Boykin et al, 2012) method. One advantage of PDC over other measures of directionality, such as amplitude cross-correlations (e.g., Adhikari et al, 2010), is that PDC allows for testing the possibility of simultaneous bi-directional communication, which is particularly likely to occur between two reciprocally connected areas, such as the ACC and VTA (Carr and Sesack, 2000;Narita et al, 2010).…”