Summary

Collectively, humans now know more about behavior than at any point in history. The pace of scientific discovery is unprecedented, with new clinical trials and experimental research being published every day. Yet despite these advances, the behavioral sciences—the social and biological sciences concerned with the study of behavior—face substantial challenges. Inconsistent use of terms and classification systems makes it challenging to integrate findings from individual studies and in turn to cumulatively build bodies of knowledge even in domains that are consistently studied. Furthermore, knowledge generated by behavioral science research is not efficiently translated for the consumers who will apply it to benefit individuals and society. The gap between what is known and the capacity to act on that knowledge has never been larger, and it continues to grow.

Ontologies provide a way to address these and other challenges in the behavioral sciences. Scientists use the word ontology to refer to efforts to structure and manage the ways in which they formally describe the entities in their discipline. Ontologies build on the philosophical notion of classifying ideas, particularly those related to the nature of existence. A widely accepted definition of a scientific ontology is a formal, explicit specification of a shared conceptualization: a commitment to using a precise, agreed-upon set of terms and relationships to represent a domain. This clear semantic specification is key to addressing the behavioral sciences’ challenges and accelerating scientific progress.

The National Academies of Sciences, Engineering, and Medicine formed a committee to study ways to improve the development and use of ontologies in the behavioral domain with support from four divisions of the National

Institutes of Health (the Office of Behavioral and Social Sciences Research, the National Institute on Aging, the National Library of Medicine, and the National Cancer Institute), the National Science Foundation, the American Psychological Association, the Association for Psychological Science, and the Federation of Associations in Behavioral and Brain Sciences. The committee appointed to conduct the study—which included experts in medicine, population health, psychology, psychiatry, biobehavioral sciences, biomedical informatics, neural and cognitive science, library and information science, the history and philosophy of science, computer science, and bioengineering—was directed to review the relevant literature, as well as example ontologies, to identify advantages and obstacles to the further development of behavioral ontologies, to identify recommended approaches to strengthening ontologies, and to offer conclusions and recommendations for advancing behavioral ontologies.

The committee focused its attention on one segment of the very broad terrain of the behavioral sciences, the domain of mental health, in order to look deeply at the literature, example ontologies, and issues in context. We explored four basic questions:

• Why do ontologies matter?
• What exactly are ontologies?
• How do ontologies facilitate advancement in the behavioral sciences?
• How can the engineering of ontologies in the behavioral sciences be strengthened?

THE BENEFITS OF ONTOLOGIES

A wide variety of stakeholders rely on the knowledge created by the behavioral sciences. Just in the domain of mental health, stakeholders include scientists and clinicians who provide educational, behavioral, social, and psychological interventions, as well as educators, health care practitioners, policy makers, and patients. The quality of the care provided to the millions of people who experience mental disorders depends both on the availability of relevant research and on the capacity of clinicians to distill relevant information from the massive volume of research published every year. The scientists who produce the research seek to test and reproduce their findings and integrate them with other knowledge. The resulting knowledge must be synthesized, generalized, and disseminated so it can be applied. Without formal ontologies, all of these functions are more difficult than they need to be.

The absence of formal ontologies—shared understanding of the concepts and phenomena being studied—also undermines the research itself. Scientists’ work is shaped by their understanding of the concepts and entities they are studying and how they are categorized, decisions about ways

to measure the phenomena of interest accurately, and decisions about what is and is not germane to their research investigations. Progress in the behavioral sciences has been hindered by the use of different terms or descriptions for the same underlying entity or condition; the use of the same term for different entities or concepts; the use of different, poorly correlated measures for the same entity; and the use of measures whose relationship to the phenomena they are measuring is not well understood. A lack of ontological clarity makes it difficult to synthesize, replicate, and generalize research findings. A key consequence is that it is difficult to build on existing knowledge, which then leads to challenges for retrieving and acting on research.

By establishing shared terms for the concepts and phenomena of interest within a particular domain and a classification of those entities, ontologies make key scientific functions possible. A clear and explicit ontology allows scientists to be precise about what they are studying and how they think about the domain in which they are working, and about the relationships among concepts, including how they are classified. By articulating a shared conceptualization of the phenomena of interest, an ontology allows scientists to communicate about ideas among themselves and with the rest of the world.

Shared ontologies help researchers to integrate their data with data developed by others, or to perform secondary analysis of online datasets. Shared ontologies make it possible for computer systems to exchange scientific claims about theories and data, thereby allowing for the development of intelligent systems and survey instruments that acquire and process data in standardized ways. Perhaps most important, ontologies make scientific research much more readily accessible not only to scientists but also to consumers, including patients searching for information relevant to their health and to practitioners who offer health services. In addition to offering scientists the direct benefits of formal specifications of entities and relationships, ontologies support scientific work indirectly, providing a foundation for scientists’ efforts to develop cumulative knowledge bases, make predictions, and develop causal explanations.

STRENGTHENING EXISTING ONTOLOGIES IN THE BEHAVIORAL SCIENCES

The committee identified numerous valuable efforts to bring ontological clarity to behavioral science but found comparatively few well-developed behavioral ontologies. A number of ontological systems that enumerate essential entities in the discipline have been developed for specific purposes and do not necessarily meet the definition of ontology. Many existing efforts have been isolated, and it appears that their adoption has been constrained. Moreover, the developers of behavioral ontologies appear to operate primarily on their own in identifying or developing the models and practices that might best suit their particular needs.

To better understand the path forward for the behavioral sciences, the committee examined the two basic components of the process of engineering formal and clear ontologies: the socio-cognitive practices and the computational tools that support the process. The socio-cognitive practices involved in creating and editing an ontology and adapting it over time require intensive human involvement. Computer tools can bring extremely valuable efficiency to the development, maintenance, and editing of ontologies, but they can never stand in for the human understanding, ingenuity, and social perceptions that go into the development and use of ontologies.

Taking advantage of opportunities to strengthen ontologies in the behavioral sciences will require attention to the practical challenges of supporting the work required. There are only a few examples of ontology development efforts in the behavioral sciences that have endured. The primary—perhaps the most important—reason for this situation is that the development and the maintenance of ontologies are both time consuming and expensive. There is no substitute, in pursuing such an effort, for the human time and intellectual effort needed, not only for the initial effort but also on an ongoing basis, as ontologies need be to continually evaluated and updated. Despite the many efficiencies afforded by computer technology, developing an ontology for any area of science is painstaking. Particularly within the behavioral sciences, there has been a lack of sustainable resources.

On the basis of these findings, the committee reached 10 conclusions:

1. Classification systems in the behavioral sciences that serve valuable ontological purposes lie on a continuum of semantic specification.
2. The classification systems that currently are widely used in the behavioral sciences do not have formal semantics, and therefore they do not readily provide opportunities to support automated reasoning and other artificial intelligence applications.
3. Ontological systems with the most formal semantic specification are not necessarily superior to others, but they offer the greatest opportunities for accelerating the behavioral sciences through the use of artificial intelligence. The most important characteristic of an ontological systems is that the level of formal specificity fits the intended purpose of the ontological system.
4. By establishing a controlled vocabulary of shared terms for the concepts and phenomena of interest within a particular domain and a classification of those entities, ontological systems have three primary benefits: opportunities to improve outcomes (care and services), infrastructure to support the mechanics and application of contemporary scientific research, and enhanced capacity to expand scientific knowledge.

5. Valuable ontological systems and related tools exist and are supporting research in the behavioral sciences. However, many of these efforts have been isolated, resources to support them (including training and education) have been limited, and the developers of ontological systems are mainly on their own to identify or develop the models, tools, and approaches that might best advance research and practice.
6. Ontology engineering rests on two foundations: socio-cognitive functions and the use of computational tools that support the process.
7. To provide the intended benefits, an ontology should be logically sound, valid, and usable by a diverse range of stakeholders.
8. For ontology engineering to progress in the behavioral sciences, sustained resources and specific actions and processes are needed for discovery, capacity building, and the promotion of practices and processes.
9. Ontology development and use has the potential to move behavioral science forward from a domain in which research is generally siloed and the data and results are often incompatible to one in which the evidence is searchable and more easily integrated and in which computer technology is leveraged.
10. Because there is no existing funding mechanism for the development and maintenance of ontological systems and the tools that support them in the behavioral sciences, sustained public and private support for the long-term development, dissemination, and maintenance of ontologies and related tools are needed.

RECOMMENDATIONS

The committee recognizes that developing and using ontologies may require complicated tradeoffs. It is possible that too much emphasis on ontological rigor could hinder originality or discourage the unorthodox thinking that has led to major scientific advances. Yet, although existing ontological systems have served valuable purposes, taken together they have not exploited the large potential for ontologies to accelerate advancement and application of behavioral research. It may be that some domains of the behavioral sciences have more to gain from a focus on ontology development than others. The committee expects that increased use of ontologies in the behavioral sciences is likely to involve different and sometimes parallel ontologies. Especially in the near term, “ontologic pluralism,” in which competing or overlapping ontologies exist but are connected to one another through formal mapping, is both inevitable and desirable.

The committee does not believe that ontologies can be expected to develop organically without encouragement: that approach has yielded the current situation. Therefore, we focused on ways to expand available

resources and incentives, stimulate grassroots ontology development, and coordinate efforts, with the aim of pushing for ontologies to be a higher priority in behavioral science research. An infrastructure is needed to coordinate and support this effort. Agencies of the federal government are the entities best positioned to provide the coordination and resources needed for this work, so we direct three of our recommendations to the National Institutes of Health, the National Science Foundations, and other federal agencies. We direct the other three to professional organizations.

Professional organizations and publishers also have a key role to play, and we direct three recommendations to such organizations. We call out specific organizations in part because they have broad reach, but we hope that similar organizations will also take part in the community building necessary to develop and encourage understanding of what ontologies can offer in the behavioral sciences.

The committee’s conclusions and recommendations build on what has been accomplished through centuries of attempts to synthesize what is known, as well as decades of research on human behavior. The actions recommended have the potential to make that knowledge more efficiently retrievable and actionable by a wide diversity of stakeholders. The committee hopes to contribute to improvement in the way knowledge is accumulated and synthesized in behavioral science and related fields. Ultimately, better communication within the scientific community and between scientists and knowledge consumers will improve the science of behavior, the way it is disseminated, and its capacity to ameliorate and prevent suffering. This report is focused on the behavioral sciences, but most of the issues discussed here would apply in other domains as well. The committee hopes these insights will be of use in advancing science generally.