BRAIN Initiative: Theories, Models and Methods for Analysis of Complex Data from the Brain (R01 Clinical Trial Not Allowed)

The Theories, Models and Methods (TMM) program provides support for the development and validation of innovative and rigorous theories, models and methods as tools that will advance a quantitative and predictive understanding of brain function across multiple scales, including behavior. Priority will be given to projects that develop novel capabilities for analyzing, integrating, and interpreting the large-scale, complex data emerging from the BRAIN initiative and related efforts, which includes cell-type specific physiological, anatomical, connectivity, and behavioral data.

Applications to this NOFO should focus on the development of fundamentally new or significantly advanced theories of brain function; mechanistic and/or predictive models of neural circuit activity and behavior grounded in empirical data; and/or novel computational or statistical methods for analyzing neural and behavioral datasets. 

TMM tools for analyzing brain activity must leverage data with cellular and sub-second temporal resolution (e.g., single-unit recordings, cellular imaging, connectomics) OR must integrate information across multiple, clearly defined temporal scales (e.g., from synaptic events to learning). Approaches relying solely on non-invasive, low-resolution signals (e.g., scalp EEG, fMRI BOLD) must be directly integrated with and constrained by cellular/circuit-level data. TMM tools for analyzing behavior must incorporate neural data AND must span multiple relevant temporal scales, capturing dynamics pertinent to the neural processes and behavior under investigation. Proposed experimental work must be limited to model parameter estimation and/or testing the validity of the TMM tools being delivered. 

It is expected that the TMM tools developed under this NOFO will be made widely available to the neuroscience research community for their use and modification. TMM projects are required to collaborate with a cohort of end users to provide user feedback. 

Specific topics of interest include, but are not limited to:

Theories of brain function

Development of predictive, mathematically-grounded theories explaining how behavior arises from neural structure, circuit dynamics, computation, cognition, and environmental variables. Examples include:

• Theories of embodied computation that anchor the neural representation of sensory, cognitive, and motor variables to an individual/animal’s ongoing interactions with the environment through dynamic, moment-to-moment, circular, and iterative processes.
• Theories that bridge multiple scales of spatial organization (e.g., molecular, synaptic, cellular, circuit, systems) and/or temporal dynamics (e.g., milliseconds to lifetimes) to generate testable predictions of brain-behavior links or cognitive function.
• Theories linking circuit dynamics and function to specific properties of cell types or anatomical connections, identifying general rules, scaling principles, and contributions of specific circuit motifs to computation.
• Theories elucidating fundamental computational principles employed by biological neural networks, potentially drawing inspiration from or contrasting with artificial networks, but firmly grounded in biological constraints (e.g., neuronal/synaptic dynamics, connectivity patterns, metabolic limits, specific cell-type properties, learning rules).

Computational models of neural and behavioral dynamics

Development and validation of quantitative models that are mechanistically grounded, interpretable, predictive, and rigorously tested against neural and behavioral data. Examples include:

• Mechanistic, interpretable, and/or predictive models of neural dynamics, circuit function, or brain-behavior links that integrate biological details with computational principles.
• Models that integrate knowledge across multiple levels (e.g., linking behavior to neural population activity and cellular/circuit properties).
• Models of cognitive processing (e.g., sensory coding, decision-making, motor control, learning, memory) that are mechanistically grounded in identified circuit elements and dynamics, make quantitative predictions, and are rigorously tested against neural and behavioral data, potentially under ecologically relevant or challenging conditions (e.g., limited information, dynamic environments).
• Development and analysis of neural-inspired computational architectures or artificial intelligence/machine learning systems explicitly designed to gain novel insights into brain function.

Methods for complex data analysis

Development of novel computational, statistical, and analytical techniques designed to extract key insights from complex, large-scale neuroscience datasets. Examples include:

• Development of innovative and scalable computational/statistical methods for dimensionality reduction, identifying latent structure, disentangling contributing factors (e.g., sensory, motor, cognitive, state variables), extracting key dynamical features, or characterizing information flow within large, complex neural and behavioral datasets.
• Novel approaches for principled data fusion and assimilation to quantitatively integrate heterogeneous datasets (e.g., linking behavior with multi-regional activity, anatomical connectivity, and cell-type information) to infer new theories of brain function, or to constrain and validate multi-scale computational models.
• Novel statistical/signal processing methods (e.g., component analysis, graphical models, compressed sensing) to track structure in neural data and link to biophysical signals for mechanistic insights across scales.

Applications Not Responsive to this NOFO

Applications deemed to be non-responsive will not proceed to review. The following are considered non-responsive for this NOFO:

• TMM tools for analyzing brain activity that do not leverage data with cellular and sub-second temporal resolution OR that do not explicitly aim to integrate information across multiple, clearly defined temporal scales.
• TMM tools relying solely on non-invasive, low-resolution signals that are not directly integrated with cellular/circuit-level data. 
• TMM tools for analyzing behavior that do not incorporate neural data AND that do not span multiple relevant temporal scales.
• TMM tool development that is not clearly aimed at elucidating or predicting neurobiological mechanisms.
• Proposed experimental work other than model parameter estimation and/or validity testing of the TMM tools being delivered.
• TMM tools, using disease or treatment paradigms, that are not used to understand underlying behavioral and functional brain circuits.