Search

Reproducibility is a core requirement of modern scientific research. For computational research, reproducibility means that code should produce the same results, even when run on different systems. A standard approach to ensuring reproducibility entails packaging a project's dependencies along with its primary code base. Existing solutions vary in how deeply these dependencies are specified, ranging from virtual environments, to containers, to virtual machines. Each of these existing solutions requires installing or setting up a system for running the desired code, increasing the complexity and time cost of sharing or engaging with reproducible science. Here, we propose a lighter-weight solution: the $\texttt{Davos}$ package. When used in combination with a notebook-based Python project, $\texttt{Davos}$ provides a mechanism for specifying the correct versions of the project's dependencies directly within the code that requires them, and automatically installing them in an isolated environment when the code is run. The $\texttt{Davos}$ package further ensures that those packages and specific versions are used every time the notebook's code is executed. This enables researchers to share a complete reproducible copy of their code within a single Jupyter notebook file.

Identifying stimulus-driven neural activity patterns is critical for studying the neural basis of cognition. This can be particularly challenging in intracranial datasets, where electrode locations typically vary across patients. This chapter first presents an overview of the major challenges to identifying stimulus-driven neural activity patterns in the general case. Next, we will review several modality-specific considerations and approaches, along with a discussion of several issues that are particular to intracranial recordings. Against this backdrop, we will consider a variety of within-subject and across-subject approaches to identifying and modeling stimulus-driven neural activity patterns in multi-patient intracranial recordings. These approaches include generalized linear models, multivariate pattern analysis, representational similarity analysis, joint stimulus-activity models, hierarchical matrix factorization models, Gaussian process models, geometric alignment models, inter-subject correlations, and inter-subject functional correlations. Examples from the recent literature serve to illustrate the major concepts and provide the conceptual intuitions for each approach., Comment: Forthcoming chapter in "Intracranial EEG for Cognitive Neuroscience"

Major depressive disorder is a common and disabling disorder with high rates of treatment resistance. Evidence suggests it is characterized by distributed network dysfunction that may be variable across patients, challenging the identification of quantitative biological substrates. We carried out this study to determine whether application of a novel computational approach to a large sample of high spatiotemporal resolution direct neural recordings in humans could unlock the functional organization and coordinated activity patterns of depression networks. This group level analysis of depression networks from heterogenous intracranial recordings was possible due to application of a correlational model-based method for inferring whole-brain neural activity. We then applied a network framework to discover brain dynamics across this model that could classify depression. We found a highly distributed pattern of neural activity and connectivity across cortical and subcortical structures that was present in the majority of depressed subjects. Furthermore, we found that this depression signature consisted of two subnetworks across individuals. The first was characterized by left temporal lobe hypoconnectivity and pathological beta activity. The second was characterized by a hypoactive, but hyperconnected left frontal cortex. These findings have applications toward personalization of therapy.

Major depressive disorder is a common and disabling disorder with high rates of treatment resistance. Evidence suggests it is characterized by distributed network dysfunction that may be variable across patients, challenging the identification of quantitative biological substrates. We carried out this study to determine whether application of a novel computational approach to a large sample of high spatiotemporal resolution direct neural recordings in humans could unlock the functional organization and coordinated activity patterns of depression networks. This group level analysis of depression networks from heterogenous intracranial recordings was possible due to application of a correlational model-based method for inferring whole-brain neural activity. We then applied a network framework to discover brain dynamics across this model that could classify depression. We found a highly distributed pattern of neural activity and connectivity across cortical and subcortical structures that was present in the majority of depressed subjects. Furthermore, we found that this depression signature consisted of two subnetworks across individuals. The first was characterized by left temporal lobe hypoconnectivity and pathological beta activity. The second was characterized by a hypoactive, but hyperconnected left frontal cortex. These findings have applications toward personalization of therapy.

The scale of functional magnetic resonance image data is rapidly increasing as large multi-subject datasets are becoming widely available and high-resolution scanners are adopted. The inherent low-dimensionality of the information in this data has led neuroscientists to consider factor analysis methods to extract and analyze the underlying brain activity. In this work, we consider two recent multi-subject factor analysis methods: the Shared Response Model and Hierarchical Topographic Factor Analysis. We perform analytical, algorithmic, and code optimization to enable multi-node parallel implementations to scale. Single-node improvements result in 99x and 1812x speedups on these two methods, and enables the processing of larger datasets. Our distributed implementations show strong scaling of 3.3x and 5.5x respectively with 20 nodes on real datasets. We also demonstrate weak scaling on a synthetic dataset with 1024 subjects, on up to 1024 nodes and 32,768 cores.

My work examines how the brain acquires, stores, and retrieves information. I first present a theoretical model of the retina, and use the model to explore how the design of sensory systems affects our ability to make inferences about the physical world. I next present three analyses of electrocorticographic recordings taken as human neurosurgical patients participated in experimental cognitive tasks. In the first analysis, I measure the relation between single-neuron spiking and local field potentials, which reflect the aggregate activity of large populations of neurons. In the second analysis, I ask how the brain represents the meanings of individual words as they are studied and remembered. In the third electrocorticographic analysis, I address the question of how our brains retrieve memories of past experiences.