Our brain’s functions and dysfunctions arise from neural activity at multiple spatiotemporal scales, from small-scale spikes of individual neurons to large-scale network activity measured through local field potentials (LFP) and electrocorticogram. Thus, developing new algorithms that can simultaneously model and decode multiple scales of activity is important for understanding neural mechanisms. Moreover, designing model-based algorithms for closed-loop control of neural activity with stimulation input is critical for establishing functional connectivity in neural circuits and devising precisely-tailored therapies for neurological disorders. We discuss some of our recent work on developing algorithms for modelling, decoding, and control of multiscale neural activity. We first present a multiscale dynamical modelling framework that identifies a unified low-dimensional latent state from hybrid spike-field activity. We show that the framework can combine information from multiple scales of activity recorded from monkeys, and model the different time-scales and statistical profiles across these scales. We then demonstrate that this framework allows us, for the first time, to decode mood variations over time from distributed multisite human brain activity. Further, we identify brain sites that are most predictive of human mood states. Finally, we present a new control-theoretic system-identification approach to characterize brain network dynamics (output) in response to electrical and optogenetic stimulation (input) within our dynamical model. To collect optimal input-output data for model estimation, we design a novel input waveform—a pulse train modulated by binary noise (BN) parameters such as pulse frequency—that we show is optimal for system-identification and conforms to clinical safety requirements. We apply the BN electrical and optogenetic stimulation waveforms in human and monkey experiments, respectively. We show that our estimated models can accurately predict both the dynamic and the steady-state neural response to stimulation input. These results show the feasibility of inferring complex brain states from distributed multiscale activity and identifying predictive dynamic models of neural response to stimulation. These algorithms could facilitate future closed-loop therapies for neurological disorders and help probe neural circuits.

The mouse is an important model for studying vision because of the molecular, imaging, and genetic tools that are available. However, critics decry the relevance of the mouse model because classical assessments based on spatial frequency analysis imply that its visual capacity is of low quality. But this is at odds with observed visually-mediated behaviors, such as prey capture, that are highly precise. We will introduce experimental data obtained with a new stimulus class -- visual flow patterns -- that formally approximate natural visual scenes, such as 'running through grass'. These stimuli evoke visually-mediated responses well beyond those predicted by spatial frequency analysis. Novel imensionality reduction algorithms reveal neural activity implicating (i) both feedforward and feedback computations; (ii) the separation of flow- from grating-responses (iii) different states of the brain, and (iv) challenges to 'receptive field' (rather than network) models of visual cortex. Joint work with the Stryker lab, UCSF, carried out with Luciano Dyballa and Dr. Mahmood Hoseini.

Most brain functions involve interactions among multiple, distinct areas or nuclei. For instance, visual processing in primates requires the appropriate relaying of signals across dozens of distinct cortical areas. Yet our understanding of how populations of neurons in interconnected brain areas communicate is in its infancy. Here we investigate how trial-to-trial fluctuations of population responses in primary visual cortex (V1) are related to simultaneously-recorded population responses in area V2. Using dimensionality reduction methods, we find that V1-V2 interactions occur through a communication subspace: V2 fluctuations are related to a small subset of V1 population activity patterns, distinct from the largest fluctuations shared among neurons within V1. In contrast, interactions between subpopulations within V1 are less selective. We propose that the communication subspace may be a general, population-level mechanism by which activity can be selectively routed across brain areas [joint work with Joao Semedo, Amin Zandvakili, Byron Yu, and Adam Kohn]