NIMH launched the Research Domain Criteria (RDoC) project to promote the translation of functional dimensions, as explicated in basic integrative behavioral neuroscience research, into clinical studies that can inform novel approaches to future diagnosis and treatment. The RDoC framework is organized around a matrix of domains and levels of analysis spanning genes, molecules, circuits, and behavior that is populated with specific elements. Since its initial development, the RDoC matrix elements have been updated and new/revised constructs of mental function are under consideration, but a thorough data-driven validation that broadly explores, compares, and validates the constructs within the matrix has not been performed.
Over the last 30 years, computational approaches have been successfully applied to basic neuroscience and contributed to significant advances in understanding complex brain functions at the molecular, cellular, systems and behavioral levels. This foundational knowledge now provides a platform for the integration of computational modeling into clinical research on psychiatric disorders. Computational approaches can advance translational research and treatment by providing better nosological classifications of heterogenous patient populations; mechanistic explanations for newly identified neuro-behavioral-types (biotypes); more precise treatment outcome predictions; and better stratification using multiple levels of data collected on individual subjects. The same approaches can also be applied to validate the functional relationships between constructs in the RDoC matrix that could, in turn, guide the development of more selective and precise treatments.
Data-driven approaches focused on integrating data between and within constructs can help improve the validity of RDoC constructs by pointing to ways in which constructs could be optimally merged, subdivided, or hierarchically organized. Additionally, quantitative approaches can provide data-driven definitions of constructs that reflect relationships among brain structures and states, networks, circuit dynamics, and hierarchies in the signals and behavioral and cognitive tasks. One approach, for example, would be to use multi-modal data fusion technology such as data-driven solutions based on matrix and tensor decompositions to unbiasedly classify and compare constructs while also identifying normative distributions, pathological outliers, or dimensional discontinuities or tipping points. Importantly, computational technology could help to link constructs across distinct domains of function and identify the convergent mechanisms by which constructs are inter-related (e.g., by clarifying the function of circuits that are involved in both the Negative Valence Systems domain and the Positive Valence Systems domain). Another approach would be to classify, predict, and explain developmental trajectories of processes and mechanisms across neuro-behavioral domains of function using cutting-edge models of dynamic change and accelerated longitudinal designs.
Applying data-driven approaches to validate the RDoC matrix will enable the matrix to become relational at the hierarchical level. The structure of a relational matrix will allow linkage of key information from different constructs through the use of functional features in the brain signal. This approach is expected to be useful in future precision medicine studies. Precision psychiatry will require the ability to analyze complex relationships within and between circuits that may be involved in multiple functions. Accordingly, a relational RDoC matrix that has been validated in a data-driven fashion will be extremely important for contributing to the next generation of targeted and precise interventions.
Competitive projects will investigate behavioral tasks that readily lend themselves to computational analysis, predictions, and explanations. These computationally-informed behaviors can be used to test detailed theories in which behaviors can be expressed as the outcome of specific circuit computations. Such projects can be divided in two classes: well-defined theory-driven projects and more exploratory data-driven projects.
Well-tested theoretical models for some of these computationally-informed behaviors already exist. For instance, in choice behavior, choice relies on valuation of different options, and how this valuation changes over time and with experience; therefore, changes in choice behavior can be traced back to changes in the underlying learning and decision-making circuit mechanisms. The development of a new computationally-informed behavior requires a theoretical framework describing how computational processes at the circuit level regulate behavior. It is important to consider algorithmic or computational accounts of behavior prior to determining whether explanations of behavior produced by mechanistic studies are sufficient to explain a given behavior. This theoretical framework needs to be accurate enough to make quantitative predictions for the behavior of interest, including statistical variability due to neuro-behavioral dynamics or continuously-valued measurements that can be used to estimate critical parameters of the underlying process. The richer the behavioral measure is (in terms of parameters that have been collected, analyzed, and modeled, e.g., reward sensitivity, learning rate, accuracy, response time, etc.), the more information it provides regarding the underlying process of interest. These approaches can be useful where we have an underlying theory or a mechanistic hypothesis to test for a particular functional construct.
In more exploratory projects, where it is unclear which features of the neuro-behavioral signal are more relevant, unbiased approaches, such as having an algorithm perform a classification, may be used to classify and predict neurobehavioral signals, to break down the behavior into its basic components and determine which neural features are better at explaining and predicting each of these components. In this way, computational methods can be used to classify both behavior and neural signal and to validate constructs in a data-driven fashion.
Purpose and Objectives
The specific goal of this Funding Opportunity Announcement (FOA) is to support projects that will use theory- or data-driven computational approaches to evaluate and optimize the validity of RDoC constructs. These projects will enable integration of theory, modeling, simulation, and analysis in RDoC-oriented experiments. Applications are encouraged to focus on established RDoC constructs; however, alternative constructs may be included if they satisfy the criteria for RDoC constructs: (1) evidence for a functional behavioral or psychological construct, (2) evidence for a neural system or circuit that plays a major role in implementing the function, and (3) a putative relationship to some clinically significant problem or symptom (e.g., anhedonia, hallucinations, social cognition). In addition, constructs should have an appropriate “grain size” (i.e., a granularity that facilitates the study of relationships among measures from the various levels of analysis), in other words, the functional aspect of the construct should be neither too broad nor too narrow to find meaningful relationships with its biological measures. The current RDoC constructs were devised with these criteria in mind, and so can serve as useful exemplars.
Examples of research projects that might be submitted under this FOA include, but are not limited to:
- Perform unbiased data-driven validation of existing constructs that may involve merging, subdividing, or hierarchically organizing them by integrating data between and within constructs. The results of such studies may indicate that no changes to existing constructs and their organization are needed, but these studies will provide a better understanding of the relationships between constructs;
- Provide a data-driven definition of a construct that involves structural and functional information regarding how brain states, networks, circuit dynamics, and hierarchies in the signals relate to outputs from task-based assays;
- Use multi-modal data fusion technology (e.g., data-driven solutions based on matrix and tensor decompositions) to unbiasedly classify and compare constructs while identifying normative distributions, pathological outliers, or dimensional discontinuities or tipping points;
- Develop computational models able to link constructs across distinct domains of function and identify the convergent mechanisms by which constructs are inter-related;
- Use accelerated longitudinal designs (with particular emphasis on development and critical periods) and cutting-edge computational models to classify, predict, and explain developmental trajectories of processes and mechanisms across neuro-behavioral domains of function.
Studies proposed under this FOA must include (all bullets must be met):
- At least two constructs; these may be existing RDoC constructs or constructs that do not appear in the matrix but meet the criteria provided above.
- Multiple behavioral tasks per construct, for convergent validity. Using distinct tasks will enable a broader, more dynamic, and complex definition of each construct.
- At least two levels of analysis (one of the levels of analysis must be a brain-based measure).
In accord with RDoC’s focus on integrative science and convergent and divergent validation of constructs, each construct may be characterized using up to seven levels of analysis: behavior, self-report, physiology, circuits, cells, molecules, and genes (see Report of the National Advisory Mental Health Council Workgroup on Genomics). Independent variables may be selected from any of the levels of analysis but must include brain-based measures. The same levels of analysis should be assessed across all the constructs to allow direct comparisons across constructs. The RDoC matrix also includes a “paradigms” column which provides examples of methods for collecting data using one or more levels of analysis but use of these specific paradigms is not required.
Some tasks and measures for RDoC constructs are reviewed in the report of the National Advisory Mental Health Council Workgroup on Tasks and Measures for RDoC . This report serves as a resource to help inform the selection of tasks; use of other tasks and measures that are relevant to the constructs and not reviewed in the report is permitted.
This FOA allows mechanistic clinical trials for studying the characteristics of neurobehavioral mechanisms or validating novel patient classification schemes (e.g., predicting differential treatment response). Additional information about the definition of mechanistic trials is provided in NOT-MH-18-004 . Applications proposing clinical trials to evaluate the efficacy of novel interventions may only be submitted under one of the NIMH clinical trials FOAs.
Examples of studies that are not responsive to this RFA and will not be reviewed include the following:
- Studies focused exclusively on an established diagnostic category or categories without consideration of phenotypic and mechanistic heterogeneity
- Studies which do not include measurement of multiple levels of analysis for each construct (one of the levels of analysis must be a brain-based measure)
- Studies which do not include a well-integrated computational and experimental component
- Studies which do not involve human subjects
Potential applicants are encouraged to review NIMH’s research priorities in the field of Computational Psychiatry (https://www.nimh.nih.gov/about/organization/dtr/adult-psychopathology-and-psychosocial-interventions-research-branch/computational-psychiatry-program.shtml) and the material at the RDoC website (http://www.nimh.nih.gov/research-priorities/rdoc/index.shtml). The RDoC website provides a detailed description of the background and rationale for RDoC, publications authored by members of the RDoC workgroup; the matrix of exemplar domains, constructs, and levels of analysis; links to prior funding announcements; summaries of grants funded under prior funding announcements; and summaries of the workshops convened to generate consensus construct specifications
Deadlines: December 3, 2018; October 20, 2019; October 20, 2020 (letters of intent); January 3, 2019; November 20, 2019; November 20, 2020 (full proposals)
URL: https://grants.nih.gov/grants/guide/rfa-files/RFA-MH-19-242.html
Filed Under: Funding Opportunities