The experimental use of antibodies that have not been validated for context-specific use frequently misdirects biomedical research. Experimental results that derive from the use of inadequately validated antibodies are estimated to waste over $1 billion annually in the United States alone and to consume millions of animal and human biological samples in experiments whose conclusions may be unreliable. Community validation frameworks, reporting standards, and independent characterisation initiatives have made important progress, and multi-stakeholder coordination efforts are emerging. However, the research community lacks a formally developed, consensus-based action plan that specifies what each stakeholder group should do, by when, and with what priority. We conducted a modified Delphi study with international experts representing academic researchers, scientific publishers, research funders, antibody manufacturers, and institutional research leaders to develop actionable recommendations for improving antibody validation, selection, and reporting practices. Thirty-two participants rated 33 proposed actions on effectiveness and feasibility using 9-point scales, with consensus assessed using the RAND/UCLA Appropriateness Method. Over two rounds, the panel achieved consensus on 15 items as both effective and feasible for implementation by 2030. These spanned institutional actions (training in antibody validation, integration into research integrity frameworks, support for local expertise networks), funder actions (dedicated validation budgets, grant application requirements, endorsement of community reporting standards), publisher actions (complete antibody reporting packages, clear validation standards), manufacturer actions (assignment of unique identifiers at source), and cross-stakeholder coordination (a shared roadmap for improvement). An additional 15 items were rated as effective but with uncertain feasibility, reflecting a consistent pattern in which the panel agreed on the value of proposed interventions but expressed reservations about realistic implementation timelines. One item was rejected by the panel with concerns around effectiveness and feasibility. Participants described four interconnected barriers to progress: diffuse ownership of the problem across stakeholders; market dynamics that inadequately reward antibody quality; difficulty justifying investment when returns are distributed across the research system; and coordination challenges among actors with different incentive structures. These barriers are addressable through coordinated action, and the findings complement existing technical and data-sharing initiatives by providing the structured, expert-informed policy framework to help translate awareness of the problem into concrete practice and policy changes.

Study design and ethical approval

We conducted a modified Delphi study to develop expert-informed recommendations for improving antibody validation, selection, and reporting practices in biomedical research. The Delphi method is a structured consensus-building approach that systematically collects and synthesises expert opinion through iterative questionnaires with controlled feedback [19,20]. This study was approved by the University of Leicester College of Life Sciences Research Ethics Committee (reference number [2025-1122-5743]) and conducted in accordance with the General Data Protection Regulation (GDPR, 2018). All participants provided written informed consent before participation. Data collection and management were conducted using the Clinvivo platform (https://www.clinvivo.com; Clinvivo Ltd, London, UK), a secure GDPR-compliant system designed specifically for Delphi studies.

Participant recruitment

We identified experts from five key sectors involved in antibody use and research governance: academic biomedical research, scientific publishing, research funding, antibody manufacturing, and institutional research leadership. Experts were defined as individuals who are representative of their profession, have the power to implement findings, or who would not be challenged as experts in their field [21]. Thirty-eight experts were identified through networks based on their expertise in antibody validation, editorial roles at major biomedical journals, leadership positions at research funding organisations, executive roles in commercial antibody production, or responsibility for research integrity policies at academic institutions. Tailored recruitment emails were sent to each stakeholder group. Participants were informed that the study aimed to develop a consensus-based action plan to improve antibody validation and reporting across the research ecosystem, and that their responses would be analysed together as a single, non-identifiable expert panel to inform policy recommendations for journals, funders, institutions, and manufacturers.

To assemble a panel spanning these sectors, we drew on the leadership and networks of established expert consortia, each asked to identify appropriate contributors from their community: the Antibody Society (representing the antibody industry and its user community), YCharOS (an independent antibody-validation consortium), the NC3Rs (a research funder with an established antibody-reproducibility programme), and the UK Reproducibility Network (providing access to institutional research leadership). A senior contact within each body was tasked with nominating relevant experts, who were then invited by the study team; these same convening bodies had contributed to an earlier multi-stakeholder meeting on antibody reproducibility [22]. Recruitment through recognised convening bodies has limitations, in that it reflects the membership and reach of those networks and may under-represent experts outside them; however, in the absence of any defined sampling frame of antibody experts, it offers a transparent and practical means of assembling a credible cross-stakeholder panel.

The Delphi process

The Delphi process consisted of iterative rounds of anonymous questionnaire completion with structured feedback. Initial draft statements for Round 1 were developed from two sources: recommendations emerging from a stakeholder meeting convened by the National Centre for the Replacement, Refinement and Reduction of Animals in Research (NC3Rs) in February 2024 [22], and themes identified through focus group discussions (and validated with a survey) with antibody-using researchers conducted in the summer of 2023 [14]. These sources yielded 23 proposed actions spanning four domains: journal and publisher policies (n = 9 items covering reporting requirements, validation transparency, and editorial capacity building), research funder actions (n = 9 items covering grant requirements, validation funding, and manufacturer engagement), institutional and educational interventions (n = 3 items covering training, ethics frameworks, and local champions), and cross-stakeholder coordination mechanisms (n = 2 items covering roadmaps, shared infrastructure, and benchmarking initiatives). Each item described a specific action that one or more stakeholder groups could adopt to improve antibody practices (S6 Text). In Round 1, participants were given the opportunity to nominate additional items not already covered in Round 1 for inclusion in Round 2.

Round 1 was conducted from August 12 to September 12, 2025, with two deadline extensions to maximise response rates. Participants received weekly email reminders from Clinvivo, and the research team sent a final reminder on the originally scheduled closing date. Round 2 was conducted from November 6 to November 23, 2025, following the same reminder protocol. The decision to proceed to additional rounds was based on pre-specified criteria: lack of consensus, persistence of significant disagreement requiring resolution, or evidence that additional rounds would meaningfully shift ratings (e.g., based on qualitative feedback).

Rating system and assessment of consensus

Participants rated each proposed action on two 9-point rating scales. The effectiveness scale permitted panellists to rate how meaningfully they thought the item would improve the selection, validation, or reporting of antibodies in research (1 = extremely ineffective, 9 = extremely effective). The feasibility scale permitted panellists to rate how realistically they thought items could be adopted across relevant settings by 2030 (1 = extremely infeasible, 9 = extremely feasible). Participants could select “don’t know” if an item fell outside their area of expertise. “Don’t know” responses were treated as non-responses and excluded from the calculations; medians, the IPR and the IPRAS for each item were therefore based only on the experts who provided a numeric rating for that item, so the number of ratings contributing to these statistics varied across items. For each item, participants could provide free-text comments to share their rationale, identify potential risks or unintended consequences, or suggest alternative wordings.

To assess disagreement, we employed the RAND/UCLA Appropriateness Method [19]. For each item, we calculated the median rating, the inter-percentile range (IPR; difference between the 30th and 70th percentiles), and the inter-percentile range adjusted for symmetry (IPRAS). The IPRAS accounts for asymmetric rating distributions using the equation: IPRAS = IPRr + (AI × CFA), where IPRr is the inter-percentile range required for disagreement when perfect symmetry exists (2.35), AI is the asymmetry index (calculated as the distance between the central point of the IPR and the scale midpoint of 5), and CFA is the correction factor for asymmetry (1.5). These values of 2.35 and 1.5 were empirically derived by Fitch and colleagues [19] to best reproduce classic RAND definitions of disagreement for 9-point scales. Percentiles were calculated using Stata version 18 (StataCorp LLC, College Station, TX; RRID:SCR_012763) with the R-2 formula (Stata’s default method).

An item was classified as having disagreement when IPR > IPRAS for that item. Items were classified according to median ratings: effective/feasible (median 7–9), ineffective/infeasible (median 1–3), or uncertain (median 4–6). For medians falling on category boundaries (3.5 or 6.5) due to even participant numbers, we applied a “round down” rule, classifying 3.5 as ineffective/infeasible and 6.5 as uncertain. This meant a conservative approach was taken to describing ratings as effective or feasible. Because this rule resolves boundary medians into the lower band, it can only have reduced the number of items meeting the effective-and-feasible criterion, never increased it; the main pattern of results is therefore not dependent on a less conservative rounding choice. Qualitative free-text comments were reviewed by the research team to inform item refinement, identify new proposed actions, and develop narrative summaries for feedback to participants in subsequent rounds.

Progression decisions between rounds

Items were progressed, removed, or modified between rounds based on their ratings, the presence or absence of disagreement, or qualitative feedback. Items rated as both effective (median ≥7) and feasible (median ≥7) without disagreement were accepted as recommendations and did not require re-rating in subsequent rounds unless there were suggestions for rewording in the qualitative comments. Items rated as both ineffective (median ≤3) and infeasible (median ≤3) without disagreement were rejected and removed from further consideration. Items classified as uncertain on either dimension, or those with disagreement, were considered for re-rating in the next round, particularly if qualitative comments suggested that rewording, clarification, or presentation of divergent perspectives might shift opinions.

Between Round 1 and Round 2, panel members’ free-text comments were reviewed to identify: (1) suggestions for new action items to add to the consensus process, (2) proposed modifications to existing item wording, and (3) substantive concerns or alternative perspectives that should be shared with the panel. Items were reworded when comments indicated ambiguity or suggested semantically meaningful improvements. New items suggested by multiple participants or representing distinct actionable proposals were added to Round 2. For items with disagreement or uncertainty, we prepared feedback summaries that synthesised the range of panel perspectives to support informed re-rating. Items with substantive wording changes based on panel feedback were re-rated regardless of initial ratings, to ensure consensus reflected the revised formulation. The number of rounds was therefore not fixed in advance but determined by these rules: rounds were planned to continue while items remained in disagreement or uncertainty and movement under directed feedback remained plausible, and the process concluded once the rating-relevant items had settled or stabilised. Consensus on every item was not the objective; a range of outcomes, with items classified as effective and feasible, effective but of uncertain feasibility, or rejected, was the intended result.

Participant characteristics and response rates

Thirty-eight individuals were invited to participate in Round 1. Thirty-two completed the round (response rate: 84.2%). All Round 1 participants were invited to Round 2, and 30 completed this round (response rate: 93.8%; overall retention: 78.9% of original invitees).

The 32 Round 1 participants included 13 females (40.6%), 17 males (53.1%), and 2 who preferred not to disclose sex (6.3%). Among the 30 participants who reported their age, the mean was 51.5 years (SD = 11.3). Participants were primarily based in the United Kingdom (n = 16, 50.0%) and United States (n = 13, 40.6%), with additional representation from Canada (n = 2, 6.3%) and Germany (n = 1, 3.1%). The panel was highly educated, with 27 participants (84.4%) holding PhD-level qualifications, 2 (6.3%) holding master’s degrees, and 3 (9.4%) holding bachelor’s degrees. Participants represented diverse fields within the biomedical research ecosystem: academic biomedical research (n = 7, 21.9%), scientific publishing (n = 6, 18.8%), antibody manufacturing (n = 6, 18.8%), research funding (n = 4, 12.5%), industrial biomedical research (n = 2, 6.3%), and other fields (n = 7, 21.9%) including antibody informatics, infrastructure provision, and the charity sector. Within these sectors, participants held senior decision-making roles: 37.5% (n = 12) reported full autonomy to make decisions on behalf of their organisations, 34.4% (n = 11) made decisions with input from senior leaders, and 28.1% (n = 9) made recommendations with final decisions made by senior leaders. Detailed breakdowns of career stages within each employment sector are provided in S1 Table.

Overview of items and consensus development

A total of 33 proposed action items were rated across two rounds (Fig 1). “Don’t know” responses were uncommon. Per item they ranged from 0 to 6 of 32 (effectiveness) and 0 to 7 of 32 (feasibility) in Round 1, and from 0 to 4 of 30 and 0 to 8 of 30 in Round 2, with most items rated by the whole panel or nearly so; the few higher counts fell on items directed at a specific sector, where panellists outside that sector could appropriately abstain. Per-item response rates for every item are reported in S2 and S3 Tables.

Round 1 presented 23 items spanning four domains: journal and publisher policies (9 items, R1–R9), research funder actions (9 items, R10–R18), institutional and educational interventions (3 items, R19–R21), and cross-stakeholder coordination (2 items, R22–R23), see S6 Text for full item wording. All 32 participants rated effectiveness and feasibility for each of the 23 proposed actions. Panel ratings demonstrated strong consensus on most items, with disagreement identified on only 2 of 46 ratings (effectiveness and feasibility combined).

Following Round 1 analysis, 11 items (R1, R2, R10–R15, R19, R20, R22) achieved consensus as both effective (median ≥7) and feasible (median ≥7) without disagreement and were immediately accepted as recommendations without requiring re-rating. The remaining 12 items from Round 1 were re-presented in Round 2, including six with revised wording based on panel comments (R3, R8, R9, R17, R18, R21). Based on participant free-text suggestions, 10 additional action items (A1–A10) were added to Round 2. Round 2 thus presented 22 items for rating: 12 re-rated items from Round 1 and 10 newly proposed items.

In Round 2, disagreement only occurred on 1 of 44 ratings. Among the 22 items presented in Round 2, four achieved agreement as both effective and feasible (R3, R21, A1, A7), joining the 11 items accepted in Round 1 for a total of 15 consensus recommendations at the close of Round 2. One item (R5) shifted from uncertain in Round 1 to consensually infeasible in Round 2 (median feasibility declined from 4.0 to 3.0) and was the only item to be rated either infeasible or ineffective. Two items (A4, A5) were rated as uncertain on both effectiveness and feasibility.

Final consensus outcomes

Across the two rounds, the panel evaluated 33 proposed actions, achieving consensus on 15 items as both effective (median ≥7) and feasible (median ≥7), without disagreement (Table 1). These 15 recommendations span all four stakeholder domains and represent actions that expert panellists agreed would meaningfully improve antibody practices and could realistically be implemented by 2030.

The panel rated 17 items as uncertain on at least one dimension (Table 2). In 15 of these 17 cases, items were rated as effective (median ≥7) but with uncertain feasibility (median 4–6), indicating that while panellists believed these actions would improve antibody practices, they had reservations about realistic implementation by 2030. This pattern—high confidence in effectiveness, uncertainty about feasibility—was the dominant finding for items not achieving consensus, and is a recurring theme in the stakeholder-specific recommendations presented below. One item was rejected by the panel, with concerns around both effectiveness and feasibility (Table 3). A summary of the ratings across all items is presented in Fig 2. A consolidated summary grouping all items by stakeholder group and rating status is provided in S4 Table.

Consistent with the criteria set out in advance for progressing between rounds, the research team reviewed the Round 2 results to determine whether a further round was warranted. This decision was based on three convergent factors. First, a high proportion of items (15 of 33, 45.5%) had achieved clear consensus as both effective and feasible across the two rounds. Second, disagreement remained minimal, with disagreement being present on only one item in Round 2. Third, rating stability between rounds was high: among the 12 items re-rated in Round 2, median effectiveness ratings remained identical for 9 items (75.0%) and median feasibility ratings remained identical for 10 items (83.3%), with no item shifting by more than 1 point on either scale. The minimal rating changes between rounds indicated that further iteration would be unlikely to resolve the persistent pattern of items rated as effective but with uncertain feasibility. Full quantitative results for all items across both rounds are presented in S2 and S3 Tables, with complete qualitative commentary provided in S7 and S8 Texts.

Stakeholder-specific recommendations

The following sections present the consensus recommendations, predicted to be both effective and feasible, and items predicted to be effective but with uncertain feasibility for implementation by 2030, organised by stakeholder group. For each recommendation, we provide the panel’s quantitative assessment alongside key themes from the qualitative feedback that contextualise the ratings and inform implementation considerations. Full qualitative analysis for the 15 consensus recommendations, is presented in S1 Text, and detailed stakeholder consultation drafts including implementation options are provided in S2–S5 Texts.

Barriers to implementation

From the panel’s free-text comments, we interpret four interconnected barriers that help explain both the persistence of the antibody validation problem and the challenges of implementing solutions. These are offered as our interpretation of the qualitative material rather than as formally coded findings.

First, diffuse ownership enables collective inaction. Antibody validation failures impose costs across the research system—wasted funding, unreliable literature, consumed biological materials—but no single stakeholder bears the full cost or has authority to mandate change unilaterally. Researchers lack incentives to validate when publication and funding systems do not reward the practice; publishers hesitate to impose requirements that could divert submissions to less demanding competitors; funders may view validation as outside their mandate; and manufacturers competing on price and catalogue breadth face limited incentives to invest in validation when researchers continue to purchase on the basis of previous use and supplier reputation [14]. Panel members described a pattern of mutual deflection, with responsibility for validation perceived as belonging to other stakeholder groups rather than one’s own (S7 Text, R22). Several characterised the broader landscape as too fragmented for meaningful alignment to feel achievable. However, the panel’s endorsement of a shared roadmap (R22) and the interdependencies across recommendations point toward a countervailing dynamic: participants noted that journal requirements would create de facto mandates for manufacturers, that researcher demand would shift manufacturer behaviour, and that funder investment in infrastructure would enable practices that other stakeholders could then require.

Second, market dynamics create structural tensions. The research antibody market operates predominantly under a Research Use Only model that prioritises catalogue breadth and low unit cost over rigorous validation. Manufacturer representatives acknowledged significant economic barriers to comprehensive validation, describing the costs as prohibitive and warning that extensive validation programmes would reduce revenue without a corresponding market incentive (S7 Text, A8). At the same time, one manufacturer described their company’s ongoing commitment to validation, illustrating that the barrier is structural rather than universal. Yet the apparent low cost of inadequately validated antibodies is illusory when downstream costs are considered: failed experiments, repeated purchasing, wasted samples, and unreliable findings impose far greater costs on the research system than better validation would. Creating market conditions in which quality is rewarded—through purchasing decisions informed by independent characterisation data, journal requirements that make validation data visible, and funder signals that prioritise quality—is necessary to shift the equilibrium.

Third, resource justification challenges persist despite clear return on investment. Participants noted competing demands on limited resources, with some questioning whether funders would prioritise antibody validation given the breadth of competing demands on their budgets, even while acknowledging the value of the proposed actions (S8 Text, A10). The economic case is compelling at a system level—the estimated $1 billion annually in direct waste in the US alone [12,13], plus far greater opportunity costs—but the return is diffuse and difficult for individual stakeholders to capture. A funder investing in validation infrastructure benefits the entire community, not just its own grantees; an institution investing in training produces graduates who carry improved practices elsewhere. The benefits are real but distributed across time and across the ecosystem, and compounded by the difficulty of demonstrating negative outcomes that were prevented.

Fourth, coordination challenges reflect the scale of the problem. With more than 22,000 human protein targets, millions of antibody products, and a global research community, the coordination challenge is immense. No single authority can mandate standards, and voluntary adoption requires alignment across actors with different incentives, timescales, and resources. The panel’s findings suggest, however, that complete alignment is not a prerequisite for progress. The 15 consensus recommendations represent actions individual stakeholders can implement without waiting for others, and participants emphasised sequencing over simultaneity: several argued that funders must first act as enablers—establishing resources, infrastructure, and training—before mandates can be effective (S8 Text, A5), while others suggested that even imperfect initial standards would allow journals to demonstrate a credible commitment to improving antibody reporting practices (S8 Text, A1).

These barriers are consistent with recent multi-stakeholder commentaries. Kahn and colleagues [12] noted that the origins of the antibody crisis are ‘many and varied’ and that ‘it is unrealistic to expect a complete solution in the near future’, while emphasising that significant progress is achievable through concerted action. In earlier work by members of this study team [18], we argued that behavioural and cultural change could be supported by engaging stakeholders at every level using established frameworks for research culture change. The Delphi findings complement these analyses by providing quantitative evidence of where expert agreement is strongest and where feasibility barriers remain.

Next steps and implementation

The findings lend themselves to a programme of targeted stakeholder consultation. To support this, we have prepared separate documents for each stakeholder group—publishers, funders, institutions, and manufacturers—presenting consensus recommendations alongside implementation options derived from the panel’s qualitative feedback (S2–S5 Texts). These present options rather than prescriptions, recognising that the optimal approach will vary across organisations, countries, and contexts. Items rated as effective with uncertain feasibility are not dismissed but treated as implementation challenges to be addressed through consultation. Interdependencies across stakeholder groups are made explicit throughout, supporting coordinated rather than siloed implementation.

Several elements of the recommended infrastructure already exist. The NC3Rs Antibody Champions programme provides a model for the local champions recommendation (R21) and delivers masterclasses for doctoral training aligned with R19. The OGA Academy offers structured online training [18], the OGA Antibody Database provides searchable characterisation data, and YCharOS continues to expand its open characterisation pipeline [1]. The eLife stakeholder roadmap [12] proposed actions that align closely with many consensus items, and the RRID initiative [17] provides identification infrastructure on which several publisher-focussed recommendations depend. These resources demonstrate that the recommendations can be implemented by building on proven approaches. Notably, several recommendations are interdependent: shared infrastructure for aggregating validation data (R23) would support automated manuscript screening (R7), informed grant review (R10), and evidence-based purchasing decisions, suggesting that investment in infrastructure could unlock a cascade of downstream improvements across stakeholder groups. As an early test of how these coordinating mechanisms might work in practice, a networked antibody champions model — embedding trained local leads within institutions to deliver training and support reagent-selection decisions — is currently being piloted across UK institutions, providing an initial operationalisation of the local-champion and roadmap recommendations (R21, R22).

To illustrate how the recommendations translate into practice, consider the consensus item asking journals to establish clear standards for antibody validation and reporting that authors must meet (A1). The consensus wording is deliberately non-prescriptive, but the panel’s qualitative feedback and the accompanying publisher consultation document (S2 Text) indicate what such standards could contain. A minimal baseline would adopt the validation and reporting language already present in the MDAR framework [16], which covers RRIDs, vendor, catalogue number, clone or lot number, and dilution. Journals could strengthen this by asking authors to indicate which IWGAV-aligned evidence [15] supports each antibody-dependent conclusion, and by requiring authors to state explicitly where context-specific validation is absent. Journals could also ask authors to supply the validation data underpinning antibody-dependent conclusions, either within the manuscript or its supplementary material or by deposition in an open repository (R4, A2). Where generating new validation data is impractical, authors could be permitted to cite previously published validation evidence obtained under closely matched experimental conditions, or to reference independent characterisation resources such as the knockout-controlled datasets generated by YCharOS [1] and aggregated in the OGA Antibody Database.

Limitations

Several limitations should be considered. Participants were identified through existing professional networks, and the panel may therefore share perspectives that do not fully represent the diversity of views across the global research community. The panel was weighted toward the United Kingdom and United States, and the biomedical research ecosystem in other countries may face different challenges. Some sectors (particularly research funders, n = 4) had few representatives, meaning feasibility assessments for funder-directed recommendations may not capture the full diversity of perspectives. The high retention rate (93.8%) and low disagreement suggest a relatively cohesive panel, which may underestimate the range of views from a broader consultation.

The Delphi method captures expert opinion at a specific point in time and does not constitute empirical evidence of effectiveness. The consensus recommendations represent the panel’s assessment of what is likely to be effective and feasible, but actual effectiveness will depend on implementation context. The panel reached consensus after two rounds, which limited iterative refinement where uncertainty persisted. However, rating stability between rounds suggests that additional rounds would have been unlikely to shift results substantially. The qualitative data were synthesised narratively rather than through formal thematic analysis, consistent with RAND/UCLA practice where free-text comments inform item interpretation rather than serving as independent qualitative data. Full qualitative commentary is provided in S7 and S8 Texts.

Conclusions

This Delphi study provides cross-stakeholder feasibility ratings with an explicit implementation horizon of 2030 and also offers prioritisation across sectors. The rating of ‘effective but uncertain feasibility’ that characterises many of the proposed actions should not be interpreted as reason for inaction. It reflects an honest assessment that the pace and sequencing of change matter as much as the direction. The 15 consensus recommendations provide an evidence-based starting point for immediate action; the remaining items rated as effective represent a second wave whose feasibility is likely to increase as early implementation builds infrastructure, shifts incentives, and establishes new norms. This staged approach—enablement to expectation to mandate—was consistently endorsed by the panel and is reflected in the consultation documents (S2–S5 Texts). Together with existing technical and community initiatives, these findings provide a structured basis for coordinated multi-stakeholder action to address a persistent and costly source of unreliability in biomedical research. Despite the challenges outlined above, this study represents an important step toward building the cross-sector alignment necessary to improve antibody validation standards and research reproducibility over time.

S2 Table. Full quantitative results for Round 1 (n = 32 participants, 23 items), including median effectiveness and feasibility ratings, inter-percentile range (IPR), IPR adjusted for symmetry (IPRAS), consensus classification, and disagreement assessment.

https://doi.org/10.1371/journal.pbio.3003981.s002

(DOCX)

S3 Table. Full quantitative results for Round 2 (n = 30 participants, 22 items: 12 re-rated from Round 1, 10 new), including median effectiveness and feasibility ratings, IPR, IPRAS, consensus classification, and disagreement assessment.

https://doi.org/10.1371/journal.pbio.3003981.s003

(DOCX)