Vincent S.C. Chien,
Richard Gast,
Burkhard Maess,
Thomas Knösche
Poster presentation link: https://meet.google.com/ukt-zxcs-jsg
The poster will be presented by Vincent Chien. To facilitate discussion, please have a look at the 10-min video (
https://youtu.be/Nxf-OQmrdpE)
that guides you through our poster.Thanks!The brain is capable of recognizing repetitive acoustic patterns within a few repetitions, which is essential for the timely identification of sound objects and the prediction of upcoming sounds. Several studies found neural correlates regarding the predictability of sequence patterns, but the underlying neural mechanism is not yet clear. To investigate the mechanism supporting the fast emergence of the predictive state, we use neural mass modeling to replicate the experimental observations during the sequential repetition [1]. First, we investigated the effect of short-term plasticity (STP) to the response of a Wilson-Cowan node to a prolonged stimulus, where the node consists of an excitatory (E) and an inhibitory (I) population. In total, 27 combinations of plasticity settings were examined, where the plasticity types include short- term depression (STD), short-term facilitation (STF), and no STP, and the connection types include E-to-E, E-to-I, I-to-E, and I-to-I connections. The simulated signals that best explain the observed MEG temporal profiles (i.e., an onset peak followed by a rising curve) rely on the setting where STD is applied on E-to-E connection and STF applied on E-to-I connection. Second, with the preferred plasticity settings (i.e., STD on E-to-E and STP on E-to-I), we simulated the dynamics of a random network in response to regular (REG) and random (RAND) sequences in PyRates [2]. The simulated signals can reproduce several experimental observations, including the above-mentioned MEG temporal profiles, the predictability-dependent MEG amplitude (i.e., dependency in terms of regularity and alphabet size of the input sequence), as well as the MEG responses in the switch conditions (i.e., from REG to RAND, and from RAND to REG). Third, we used a simplified two-level network to illustrate the main mechanisms supporting such representation of predictability during the sequential repetition. The simplified network consists of nodes that are selective to sound tone (level 1) and nodes that are selective to tone direction (level 2). The simulation reveals higher firing rates of I populations level-2 nodes during REG than RAND condition, which contributes to stronger simulated MEG amplitude via I-to-E connections (Fig 1). In conclusion, we provide a possible mechanism to account for the experimental observations. First, the increased MEG amplitude is mainly due to increased inhibitory activities. Second, the effect of alphabet size is due to two forms of STP (i.e., STD on E-to-E and STF on E-to-I). Third, the effect of regularity relies on the inclusion of the 2nd-level nodes that sparsely encodes the repetitive patterns. In short, the more predictable sequence patterns cause a stronger accumulation of inhibitory activities in direction- selective areas via STP, which in turn leads to a higher MEG amplitude. This mechanism emphasizes the need for STP at each stage of the bottom-up process, whereas the involvement of top-down processes is not necessary.
[1] Barascud, Nicolas, et al. "Brain responses in humans reveal ideal observer-like sensitivity to complex acoustic patterns." Proceedings of the National Academy of Sciences 113.5 (2016): E616-E625.
[2] Gast, Richard, et al. "PyRates—A Python framework for rate-based neural simulations." PloS one 14.12 (2019).
