2. Case study: The National Academies of Science, Engineering, and Medicine report on the use of population descriptors in genomics

Race, ancestry, and ethnicity are contentious categories in human genomics (Caulfield et al. Reference Caulfield, Fullerton, Ali-Khan, Arbour, Burchard, Cooper and Hardy2009; Bliss Reference Bliss2012; Yudell et al. Reference Yudell, Roberts, DeSalle and Tishkoff2016). These categories work as population descriptors, i.e., as different classification schemes that organize samples into groups as a way of studying patterns of genetic distribution. However, the use of these descriptors raises ethical and epistemic issues. Ethically, scientists and policymakers worry that human genomics might cause harm by reifying racial categories or by disguising racial assumptions with the new terminology of ancestry (Fullwiley Reference Fullwiley2008; Cerdeña et al. Reference Cerdeña, Asabor, Plaisime and Hardeman2022; Lewis et al. Reference Lewis, Molina, Appelbaum, Dauda, Rienzo, Fuentes and Fullerton2022). Epistemically, the concern is that genetic studies use those population descriptors irreflexively and inconsistently (Caulfield et al. Reference Caulfield, Fullerton, Ali-Khan, Arbour, Burchard, Cooper and Hardy2009; Panofsky and Bliss Reference Panofsky and Bliss2017; Zhang and and Finkelstein Reference Zhang2019). It is common to find inconsistent use of racial, ancestry, and ethnicity categories in the same study (Caulfield et al. Reference Caulfield, Fullerton, Ali-Khan, Arbour, Burchard, Cooper and Hardy2009), thus creating problems for the replicability and validity of such studies.

In this context, the National Institutes of Health (NIH) sponsored a consensus study conducted by the National Academies of Science, Engineering, and Medicine, gathering a committee of 17 members with expertise from various areas of the natural and social sciences, as well as stakeholders in genetics companies and members of the biomedical community. The result was a report titled “Using Population Descriptors in Genetics and Genomics Research: A New Framework for an Evolving Field” (NASEM 2023), whose goal was to review and assess existing methodologies, benefits, and challenges in using race, ancestry, and ethnicity in genomics research (2). Additionally, the document offers technical recommendations and best practices to guide future work in the field.

Recommendations and the suggested best practices are explicitly motivated by a set of “guiding principles” established by the committee (95). These principles are validity, reproducibility, replicability, transparency, respect, beneficence, equity, and justice. Noticeably, such principles correspond to epistemic and non-epistemic values, respectively (Elliott Reference Elliott2022; Elliott and Korf Reference Elliott and Korf2024). Some of them are meant to systematically indicate or promote empirically valid research, while others are associated with ethical and social standards and do not necessarily advance knowledge.Footnote ² The combination of epistemic and non-epistemic values is key to the recommendations proposed by the National Academies of Sciences. Together, these values would ensure that research is both ethically conducted and empirically valid. They are supposed to make research on human genomics trustworthy, engaging both researchers and other stakeholders. The report nonetheless acknowledges the possibility of conflict among values, claiming that “tension between principles may arise when researchers prioritize competing interests” (99). One example would be choosing categories that maximize data aggregation and, thus, reproducibility, but do not respect how humans would like to be classified (Lee et al. Reference Lee, Fullerton, Saperstein and Shim2019). We will come back to this point in the next section.

The brief discussion of “guiding principles” is followed by a set of recommendations and best practices that indicate which population descriptors should be adopted depending on the type of genomics research being conducted (121). While not always explicit, these recommendations and practices are supposed to result from applying the guiding principles to different research contexts. For example, the category “race” is only recommended if scientists use it as a proxy for how environmental factors (e.g., education, access to food, living conditions) impact health disparities in human populations. This category fails to meet several, if not most, principles and endangers both ethical and empirical dimensions of research. Meanwhile, the report proposes genetic similarity as a population descriptor that should be adopted in all but one type of genomic study (Coop Reference Coop2023). Genetic similarity is a quantitative measure of the genetic resemblance between individuals and can help scientists infer genetic ancestry, patterns of genetic and phenotypic diversity, among other things (NASEM 2023, 117). Hence, this descriptor can (and should) be widely used, serving as a proxy for both biological and environmental factors in medical and evolutionary contexts. Another advantage of this descriptor is avoiding the moral baggage of notions like race and ethnicity.

This case study exemplifies how values inform scientific recommendations. Scientists are aware of such values and the fundamental role they play in grounding best practices. Scholars even recognize that value trade-offs might be inevitable. However, values raise challenges for the development and the uptake of recommendations. Recommendations are meant to be applicable in concrete circumstances, and the National Academies of Science report takes several steps in this direction. For example, the report suggests actionable decision trees to guide adherence to the recommendations (191–97). Therefore, the first challenge is to develop realistic recommendations that sufficiently consider the relevant trade-offs involved. Ignoring contextual elements and the ensuing trade-offs may result in recommendations that are superfluous and unlikely to be incorporated into scientific work. In the following section we explain and illustrate this problem, showing how the report produced by the National Academies of Science falls prey to it.

3. Challenge I: Overlooking value trade-offs

Epistemic and non-epistemic values motivate the adoption of recommendations in the National Academies of Science report (2023). On the one hand, validity, reproducibility, transparency, and replicability are paradigmatic examples of epistemic values, as they are indicative of and conducive to knowledge (Elliott Reference Elliott2022, 4). On the other hand, respect, beneficence, equity, and justice correspond to moral and socio-political standards and are not necessarily associated with knowledge. While these eight values are emphasized in the report, they are not the only values that (should) influence the recommendations around population descriptors. In fact, the very discussion of recommendations in the report appeals to additional values such as precision and rigor.

Here we focus on one flagrant trade-off ensuing from the report’s recommendations: the trade-off between precision and generality. While human genetic variation is continuous, population descriptors appeal to generalized, coarse-grained categories (Lewis et al. Reference Lewis, Molina, Appelbaum, Dauda, Rienzo, Fuentes and Fullerton2022; Gannett Reference Gannett2014). These categories abstract away fine-grained differences among samples, typically lumping them into a few groups to represent broad patterns of genetic variation. Hence, humans tend to differ gradually from each other based on factors such as geographical distance, but imprecise measurements and coarse-grained descriptors might hide this fact (Lewis et al. Reference Lewis, Molina, Appelbaum, Dauda, Rienzo, Fuentes and Fullerton2022, 251). Scientists generally agree that it is more precise and more accurate to represent and convey variation as fine-grained and as continuously as possible.

Precision, alongside accuracy, is a key aspect of the reliability of measurements in science (Tal Reference Tal and Edward2020). Technically, precision can be understood as “the degree to which repeated measurements by the same procedure agree” (Isaac Reference Isaac2022, 1245). Take, for example, a person’s ancestry estimation. Such an estimate will be deemed precise insofar as it accurately captures the closeness between the predicted ancestry and an individual’s “actual” ancestry, assessed through their self-identification at the recruitment stage of a study. Therefore, to make a precise estimation of ancestry entails that the same procedure yields similar estimations for individuals who share the same or similar ancestry.

Precision is an epistemic value, i.e., it improves the overall quality of measurement in a specific scientific practice, thus allowing for better predictions and a better understanding of the phenomenon under inspection. Precision is supposed to increase the empirical adequacy and transparency of research methods and results, reducing the chances of misconceptions and erroneous interpretations of genomic research (NASEM 2023). Accordingly, in the context of predicting and discovering diseases, scientists should avoid “poor proxy” and use “reliable and precise measures” (123–28). An adequate proxy will offer precision, enabling a precise assessment of disease risk in terms of risk scores (Woodward and Kendler Reference Woodward and Kendler2023; Cavazos and Witte Reference Cavazos and Witte2021). Inadequate proxies due to too coarse-grained labels may lead to disrespect and injustice, for instance by exacerbating health disparities.

Unsurprisingly, the National Academies of Science’s report implicitly sustains that precision fosters trusting relationships between scientists and the broader lay community. For instance, precision appears when suggesting that the descriptor “race” does not have the desirable characteristics of “precision, rigor, stability, and objectivity one might hope for in a population descriptor” (NASEM 2023, 75). The report explicitly calls for scientists to use measures that are more precise and accurate than racial labels (ibid.).

Emphasis on precision, however, implies a trade-off with another epistemic value: generality. Generality—broadly defined as “the property of applying widely”—also brings epistemic benefits to scientific practice (Lehtinen Reference Lehtinen2021, 10344), such as widespread sharing of results and broader extrapolation of research findings across other populations. Here, generality refers to the extent and degree to which samples fit a classification scheme. For instance, descriptors such as “ethnicity” and “ancestry” are relatively general because they can apply to a wide variety of samples in common categories. These are known as “group labels” and enhance generality, since they can be applied to several individuals in a group as they are coarse-grained. However, this means that group labels lack the precision of fine-grained labels. Generality thus comes at the cost of enabling precise comparisons and differentiations among samples, which can be done with more precise descriptors. For example, “relatedness” or “genetic similarity” (Coop Reference Coop2023) can generate fine-grained classifications which are more precise but less general.

This trade-off creates a problem for the scientific community. On the one hand, scientists strive for as much precision as possible. The report suggests that precision contributes to the other epistemic values, leading to empirically valid and trustworthy research (NASDEM 2023, 95). On the other hand, generality is another key epistemic value in genomics. It facilitates data aggregation, data sharing, cross-study referencing, and other core aspects of data-centric biology (Leonelli Reference Leonelli2019). Moreover, the generality of group labels facilitates collaborations within the scientific community as well as enrolment of participants in a study, since general labels result in individuals identifying more easily with the classification scheme employed at the recruitment stage. Finally, coarse-grained group labels also facilitate enrolment as they are less likely to hurt individual sensitivities that more precise labels might awaken (Panofsky and Bliss Reference Panofsky and Bliss2017).

Recommendations that focus on precision—such as those in the NASEM report—may naively ignore the trade-off with generality. They bias scientists toward the value of precision when it might be legitimate to privilege generality instead. There might be no simple solution to certain trade-offs, no “one size fits all.” In these cases, recommendations biased toward one side of the trade-off will likely suffer the problem of uptake. The scientific community might adhere poorly to such recommendations because they do not capture what scientists might actually and legitimately value in their practices. In other words, recommendations might be too unrealistic if they are premised on the assumption that scientists will (and should) embrace certain values over others.

4. Challenge II: Oversimplifying value s ynergies

Value trade-offs happen when prioritizing certain values comes at the cost of other values, indicating a tension between pursuing different values at the same time. Besides trade-offs, there are also cases when there is a positive reinforcement among values, which we call value synergiesFootnote ³ and are the object of this section. We consider how the oversimplification of value synergies can create a problem for the uptake of the recommendations.

The primary target of the proposals put forward by the National Academies of Sciences are researchers working in the field of human genomics who are expected to adhere to these recommendations (NASEM 2023, 3). An explicit goal is that these recommendations—if correctly adopted—“engender trust and drive trustworthiness of research” (5). Recommendations should therefore guide scientists in building trust with relevant stakeholders such as, for example, affected communities. The report recognizes that mistrust is a significant problem in the relationship between human geneticists and those communities (9). Among other issues, biases in data collection reinforce mistrust leading to the underrepresentation of these communities in research (Popejoy and Fullerton Reference Popejoy and Fullerton2016). Hence, recommendations are both informed by values (e.g., “guiding principles”) and meant to contribute to further values (i.e., trust and trustworthiness). The underlying idea is that the pursuit of epistemic values such as precision leads to the promotion of other values such as trust, thus illustrating what we call value synergies.

Specifically, there are three steps that illustrate the occurrence of value synergies. First, recommendations are based on a set of primary values that are openly endorsed as the “guiding principles” of the report. These values are validity, reproducibility, replicability, transparency, respect, beneficence, equity, and justice. Second, these values are endorsed because they promote an additional set of values: empirical soundness and ethical research conduct. Third, these values are positively related to trust and trustworthiness. The report explicitly describes those values as “aspects of trustworthiness” (NASEM 2023, 95). While there is no clarification of what exactly this means, one plausible interpretation is that trust and trustworthiness require empirical soundness and ethical research conduct. Thus, trust and trustworthiness of research are core non-epistemic values (or goals) reinforced by the primary values (“guiding principles”) embedded in the recommendations. In summary, the primary values are supposed to stand in synergy with the core values. Promoting the former supposedly contributes to promoting the latter.

Value synergies can be complex and opaque to scientists and other stakeholders. It is not always clear whether and when a certain value contributes to another. Additionally, synergies might vary according to context, and scientists might hold misleading assumptions about those synergies. For instance, the National Academies of Sciences’ report assumes that precision stands in a synergistic relationship with trust. The idea is that precision might contribute to, “unlock,” or “enhance” trust, even if it is not sufficient for it. This is a very plausible assumption in the context of using population descriptors. However, one must be careful when embracing this assumption because precision is not sufficient for trust. The emphasis on the precision–trust relationship might lead scientists to miss other synergies involving trust. In particular, trust might be promoted by values that stand in tension with precision, such as the value of generality. We explore this case below.

As indicated by the NASEM report, a population descriptor like “genetic similarity” is more precise than “ancestry” (section 3). The latter enables coarse-grained classifications, typically including continental-level categories such as “European,” “African,” and “Asian,” while the former enables finer-grained classifications that consider highly specific genetic differentiation. These classifications might not even conform to geographical regions and, when conforming to them, they offer transparent and detailed descriptions of what is being measured (Coop Reference Coop2023). For example, consider the ancestry-based claim “Graham has Northwestern European ancestry” (Coop Reference Coop2023, 11). If scientists adopt genetic similarity instead of ancestry, this claim will turn into something like “Graham is more similar to the Great Britain sample from the 1000 Genomes Project relative to all the other samples in the same data bank.” Given that precision is meant to foster trust and trustworthiness, the NASEM report suggests that genetic similarity (and its detailed claims about someone’s origin) tends to promote trust and trustworthiness.

However, there are foreseeable situations in which more precise population descriptors do not enhance those core values, such as in historical examples of how genomics has been conducted in minority groups, leading to the erosion of trust between stakeholders. Examples include the ongoing misuse of racial categories (Kozlov Reference Kozlov2024), the deliberate overlook of self-identification of certain groups (TallBear Reference TallBear2013), the extractivist culture in the use of genetic data (Reardon Reference Reardon2001), and the pernicious interpretation of findings to support scientific racism (Bolnick Reference Bolnick, Koenig, Lee and Richardson2008; Gannett Reference Gannett2004). All these examples reinforce that trust and trustworthiness are delicate subjects in human genomics. Indeed, sampled subjects might prefer population descriptors that reflect generality rather than precision if the former better aligns with how those subjects self-identify. For instance, Panofsky and Bliss (Reference Panofsky and Bliss2017, 72) note that over the past decades researchers have been increasingly dependent on the good will of study participants as they need to collect data from thousands of individuals. Ambiguous labelling schemes based on continental labels (which deliberately mix racial/ethnic categories) are attractive for scientists since they facilitate data acquisition and are less likely to hurt individual sensitivities with respect to classification (Panofsky and Bliss Reference Panofsky and Bliss2017). This is likely because study participants identify more easily with coarse-grained and less precise population descriptors. In situations like this, assuming a synergy between precision and trust is an oversimplification. Precision is not sufficient for trust. Furthermore, generality might be a more relevant contributor to trust than precision. The emphasis on the precision–trust relationship in the NASEM report misses that contribution.

The assumption of synergy between values has consequences for the effectiveness and uptake of recommendations. Value synergies are complex because they can be highly context sensitive, varying from situation to situation. Oversimplification happens when local variation is ignored, such as when there is a uniform treatment of study participants disregarding local and historical contexts. However, different contexts will influence how trust can and should be promoted. Some contexts might require emphasis on precision, while others might require emphasis on generality. The consequence is that, when recommendations are predicated on oversimplified value synergies, recommendations might be so oversimplified that they become ineffective even if adopted. For instance, it may be the case that enhancing precision simply makes no difference whatsoever to trust and trustworthiness of research being conducted in some historically marginalized communities. In fact, it may even be the case that recommendations aiming to promote trust might contribute to distrust. An example of this can be seen in the outcomes of the Human Genome Diversity Project, which sought to better represent human genetic diversity by including local indigenous populations (historically marginalized and under-represented in such studies). Despite good intentions, the study was stigmatized and eroded trust with such populations due to the lack of historical sensitivity. The project came across as a threat to the self-identity of those populations and failed to offer concrete benefits to them (Reardon Reference Reardon2001, Reference Reardon2005; Bliss Reference Bliss2012). Even though researchers sought more precision as a general value, it was not sufficient to enhance trust and trustworthiness of research (Reardon Reference Reardon2001).

In sum, recommendations that oversimplify value synergies will fail to foster the desired value (goal) because they may be based on a distorted understanding of how to achieve it. After all, as we have illustrated in the previous paragraph, even when precision is enhanced there are reasons to doubt the trustworthiness of research in genomics. It is very plausible to assume that precision is a contributor to trust, but this assumption should not come at the expense of acknowledging that generality can also be a contributor to trust. Contributions can be more or less relevant depending on the context. Failing to recognize the relevant contributors and synergies leads to ineffective recommendations.

5. Moving forward

So far, we have outlined two challenges to the uptake of value-laden recommendations: overlooking value trade-offs and oversimplifying value synergies. More needs to be said about each challenge and how values influence scientific recommendations. However, it is also important to consider potential strategies to overcome those challenges. In this section, we briefly assess some of these strategies.