NiSpace: Neuroimaging Spatial Colocalization Environment
Poster No:
2265
Submission Type:
Abstract Submission
Authors:
Leon Lotter1,2,3, Simon Eickhoff1,2, Juergen Dukart1,2
Institutions:
1INM-7, Research Centre Jülich, Jülich, NRW, Germany, 2Institute of Systems Neuroscience, Heinrich Heine University, Düsseldorf, NRW, Germany, 3Max Planck School of Cognition, Leipzig, Saxony, Germany
First Author:
Leon Lotter
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University|Max Planck School of Cognition
Jülich, NRW, Germany|Düsseldorf, NRW, Germany|Leipzig, Saxony, Germany
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University|Max Planck School of Cognition
Jülich, NRW, Germany|Düsseldorf, NRW, Germany|Leipzig, Saxony, Germany
Co-Author(s):
Simon Eickhoff
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Juergen Dukart
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
INM-7, Research Centre Jülich|Institute of Systems Neuroscience, Heinrich Heine University
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Introduction:
Most neuroimaging research focuses on brain-regional effect estimates. Spatial patterns across the whole brain or cortex may convey equally important information missed by regionally constrained analysis approaches. Such patterns are found in any brain map, from individual brain development over time [1], to effect size maps reflecting case-control differences on the individual [2] or group level [3]. By testing these maps for their colocalization with atlases of known biological entities (e.g., neurotransmitters, gene expression, or cortical microstructure), one can derive meaningful inferences from in-vivo human neuroimaging data. The most renowned spatial colocalization tools are JuSpace [2] and neuromaps [4]. While JuSpace focuses on GUI-based evaluation of clinical case-control differences using uni- and multivariate colocalization statistics in MNI space, neuromaps' core features include a multimodal atlas dataset, flexibility in regard to imaging spaces, and several spatial null models; with limited functionality in terms of colocalization statistics and analysis workflows.
Here, we introduce NiSpace ("NeuroImaging SPAtial Colocalization Environment"), a user-friendly toolbox integrated in the Python neuroimaging ecosystem that allows for fast, flexible, and scalable colocalization analysis.
Here, we introduce NiSpace ("NeuroImaging SPAtial Colocalization Environment"), a user-friendly toolbox integrated in the Python neuroimaging ecosystem that allows for fast, flexible, and scalable colocalization analysis.
Methods:
NiSpace employs a pipeline-style API (Fig. 1), extended by simplified command line and graphical interfaces for less experienced users. It advances available tools in that it (i) interfaces with existing multimodal human brain datasets, (ii) handles surface and volumetric imaging data types, (iii) allows for flexible and scalable analysis workflows, (iv) incorporates vectorized implementations of all currently used spatial colocalization statistics, (iv) provides permutation-based significance estimates, and (v) produces publication-ready visualizations.
User-defined input brain maps are tested for colocalization with a set of multimodal brain atlases. All brain data is transformed to a common space [4], parcellated and, if required, group difference maps [2] are calculated. Colocalizations are evaluated by treating individual parcels as "observations" and brain maps as "features". In the univariate case, every input map is correlated with every multimodal atlas, with correlation coefficients as estimates of colocalization [2]. In the multivariate setting, each input map is "predicted" from all multimodal atlases, resulting in R2 values that indicate how well a given map can be explained [1,3]. All statistics are furthermore employed in gene set enrichment analyses based on the Allen Brain Atlas [5,6]. To reduce false-positives, NiSpace provides exact p values based on permuted brain maps [7], subject groups [2], or gene sets [5].
User-defined input brain maps are tested for colocalization with a set of multimodal brain atlases. All brain data is transformed to a common space [4], parcellated and, if required, group difference maps [2] are calculated. Colocalizations are evaluated by treating individual parcels as "observations" and brain maps as "features". In the univariate case, every input map is correlated with every multimodal atlas, with correlation coefficients as estimates of colocalization [2]. In the multivariate setting, each input map is "predicted" from all multimodal atlases, resulting in R2 values that indicate how well a given map can be explained [1,3]. All statistics are furthermore employed in gene set enrichment analyses based on the Allen Brain Atlas [5,6]. To reduce false-positives, NiSpace provides exact p values based on permuted brain maps [7], subject groups [2], or gene sets [5].
·Figure 1: Workflow overview
Results:
NiSpace's core functions (Fig. 1) include:
NiSpace.fit(): The user provides volumetric, surface, or parcellated neuroimaging data from a given number of subjects (Y) along with integrated or user-defined biological/functional maps of interest or gene sets (X), and a brain parcellation.
NiSpace.transform(), .harmonize(), .compare(): If required, the parcellated data can be cleaned from covariates, harmonized across study sites, or parcel-wise compared between groups.
NiSpace.colocalize(): Colocalization estimates are calculated. A focus is laid on regression-based approaches including dominance analysis, partial least squares, and regularized regression, using within-brain distance-dependent cross validation [8].
NiSpace.permute(): Exact p values are calculated based on permuted X/Y brain maps, Y groups, or X gene sets.
NiSpace.visualize(): Adjustable plotting functions allow for unbiased interpretation and publication-ready visualization.
NiSpace.fit(): The user provides volumetric, surface, or parcellated neuroimaging data from a given number of subjects (Y) along with integrated or user-defined biological/functional maps of interest or gene sets (X), and a brain parcellation.
NiSpace.transform(), .harmonize(), .compare(): If required, the parcellated data can be cleaned from covariates, harmonized across study sites, or parcel-wise compared between groups.
NiSpace.colocalize(): Colocalization estimates are calculated. A focus is laid on regression-based approaches including dominance analysis, partial least squares, and regularized regression, using within-brain distance-dependent cross validation [8].
NiSpace.permute(): Exact p values are calculated based on permuted X/Y brain maps, Y groups, or X gene sets.
NiSpace.visualize(): Adjustable plotting functions allow for unbiased interpretation and publication-ready visualization.
Conclusions:
NiSpace is a Python toolbox with a dedicated API providing a user-friendly spatial colocalization environment that integrates a diverse range of brain atlases with all currently used analysis strategies, robust and scalable statistical testing, and results presentation.
Genetics:
Transcriptomics
Modeling and Analysis Methods:
Methods Development
Neuroinformatics and Data Sharing:
Brain Atlases 2
Workflows 1
Physiology, Metabolism and Neurotransmission :
Physiology, Metabolism and Neurotransmission Other
Keywords:
Data analysis
Neurotransmitter
Open-Source Software
Statistical Methods
Workflows
Other - Spatial Colocalization; Spatial Correlation; Imaging Transcriptomics
1|2Indicates the priority used for review
Provide references using author date format
2. Dukart, J. et al. (2021), "JuSpace: A tool for spatial correlation analyses of magnetic resonance imaging data with nuclear imaging derived neurotransmitter maps", Human Brain Mapping, doi:10.1002/hbm.25244
3. Hansen, J.Y. et al. (2022), "Mapping neurotransmitter systems to the structural and functional organization of the human neocortex", Nature Neuroscience, doi:10.1038/s41593-022-01186-3
4. Markello, R.D. et al. (2022), "neuromaps: structural and functional interpretation of brain maps", Nature Methods, doi:10.1038/s41592-022-01625-w
5. Fulcher, B.D. et al. (2021), "Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data", Nature Communications, doi:10.1038/s41467-021-22862-1
6. Hawrylycz, M.J. et al. (2012), "An anatomically comprehensive atlas of the adult human brain transcriptome", Nature, doi:10.1038/nature11405
7. Burt, J.B. et al. (2020), "Generative modeling of brain maps with spatial autocorrelation", NeuroImage, doi:10.1016/j.neuroimage.2020.117038
8. Hansen, J.Y. et al. (2021), "Mapping gene transcription and neurocognition across human neocortex", Nature Human Behavior, doi:10.1038/s41562-021-01082-z