Funding knowledgebases: Towards a sustainable funding model for the UniProt use case

1 Introduction

2 Overview of existing funding models

In this paper, twelve funding model have been identified and are described here below. They include existing funding models for data resources and facilities, as well as possible scenarios that are currently considered by various international initiatives. The list is not exhaustive, but introduces the major funding sources in the various sectors of research, with a particular focus on the life sciences. Some funding models that are currently not specifically supporting data resources are also included, as they could be implemented for data stewardship and preservation.

The twelve models can be grouped in three main categories in function of the revenue origins (Figure 1). Most of the models rely on funds coming from national budgets and allocated to research and/or infrastructure, redistributed among the applicants, institutions or services according to various rules and conditions. A second category embraces all the models dependent on user fees. Finally, there are models counting on voluntary donations and participations, or third party funding. On top of these, models exist that are a mixture of these categories, as they benefit together from national funding and commercial fees or investments, or they take advantage of funding from government bodies, industries and voluntary donation.

Figure 1. Funding models sources.

The 12 considered models are represented depending on the origin of the revenues.

1. National funding: governmental agencies fund the infrastructure directly, through non-cyclical funding programmes. For research infrastructures, funds derive directly from the domestic R&D budgets. Often, users are charged for some subscriptions or special services. Examples are:

2. Infrastructure model⁹: funding agencies pay directly for data resources as a necessary part of the research infrastructure, through a percentage of the research funding that is specifically set aside. The grants themselves are only allocated to research projects. A percentage of each grant is then retained and assigned to a budget for data stewardship, and subsequently redistributed among the relevant infrastructures, including knowledgebases. This model is similar to the National model (model 1), but in this case funding agencies are not necessarily national (they can also be private, thus with different budget constraints). The funding agencies contribute financially in proportion to the grant volume that they allocate to research. This model is not implemented yet as a funding model for life sciences knowledgebases.

3. Institutional support: universities or institutions have their own repository/data bank that is maintained through the “internal” institutional funds. Grants can be cyclic or long-term, and usage may be restricted to the institution’s members or be open to the worldwide community.

4. Research project grants: competitive cyclic research or dedicated resource grants from national funding agencies such as the NIH, the National Science Foundation (NSF) or the Swiss National Science Foundation (SNSF). They request a submission by the applicant every 3–5 years. Access is free for the user. This category includes also a few existing grants specifically conceived for databases and resources⁵. Most of the databases and knowledgebases in the life sciences are supported by these type of grants, including:

5. Content licensing/industrial support model¹⁰: requires commercial users to pay a fee for access to the data and for-profit reuse, whereas data are free for non-commercial users.

6. User subscription fees: users are charged on a time base (e.g. every month or year) or on download sizes, and they have access to the entire database. At the end of the validity, the subscription must be renewed to continue the access.

7. Value-added/asymmetrical pricing model (freemium service)⁸: a basic data set within the database is freely available to anyone. Individual scientists or companies that are willing and able to pay a higher fee can buy additional levels of service, better data access or additional tools and resources.

8. Infrastructural razor & blades¹⁴: an attractive, inexpensive or free initial offer (“razor”) encourages continuing future purchases of follow-up items or services (“blades”).

9. Public-private consortium⁷: is a mixture of funding from government bodies and industries. The funders mandate the research subjects and supporting companies do not receive priority access to data.

10. Online advertising and corporate sponsorship: corporate sponsorship is part advertising and part dealmaking —the corporation pays to support a database that provides value to its potential customers.

11. Open source volunteering (or wiki approach)¹⁵: replacing part of data curation by community participation can be attractive as it has a low cost. It depends, however, on drawing contributions from busy users. In addition, contributions tend to be sporadic, leaving many gaps. Hence it can only replace (a small) part of curation and therefore still requires funding for curation, software engineers, storage space, and operating costs.

12. Donations: philanthropic funding such as grants and donations can generate income. They partly depend on the impact on and awareness of the (user) population.

13. Mixed models: most of the knowledgebases rely on diversified multiple funding streams. This approach has the obvious advantage of increasing resilience if one of the sources disappears. Some example of databases or knowledgebases supported by a mixed model are:

The models are summarized for comparison in Table 1. Each model is described in terms of its compatibility with open access policies and its equity among the potential users and institutions, i.e. whether or not wealthier institutions or certain users are particularly favoured. Also, the forecasted stability of the models over time and the key dependency of each funding stream are indicated. Associated factors such as national/international economic situation dependency (which are obviously relevant within each model described) has been indicated only when representing the main dependency. The dependency of the funding is crucial to describe the vulnerability of the models and also needs to be taken into account when setting up a mixed model. The best funding model would combine models that are dependent on different factors.

3 Funding situation of the UniProt knowledgebase, past and present

For the purpose of this work, the UniProt knowledgebase is used as a case study. UniProt contains a reviewed collection of high-quality annotated and non-redundant protein sequences, and brings together experimental results, computed features and scientific conclusions. At present, UniProt is developed and maintained by the UniProt consortium, a collaboration between the SIB Swiss Institute of Bioinformatics, the European Bioinformatics Institute (EMBL-EBI), and the Protein Information Resource (PIR). The UniProt knowledgebase (UniProtKB) is the central resource that combines UniProtKB/Swiss-Prot and UniProtKB/TrEMBL. UniProtKB/Swiss-Prot contains sequences that have been curated by expert biocurators, while UniProtKB/TrEMBL provides sequences that have been annotated by automatic annotation systems. UniProt also includes the UniProt Reference Clusters (UniRef), a database of clustered sets of sequences from the UniProtKB, and the UniProt Archive (UniParc) that provides a complete set of known sequences, including historical obsolete sequences.

The UniProt knowledgebase is an interesting case study because it passed through various funding models, as well described in the literature^7,18,19. It started under the name of Swiss-Prot, a research project at the University of Geneva in 1986. At that time it was funded through a Swiss National Science Foundation (SNSF) research grant, which lasted until 1996, when the knowledgebase suffered a funding crisis^20,21. After negotiations with the Swiss Government, an agreement was reached with the creation of an institutional framework for the knowledgebase: the SIB Swiss Institute of Bioinformatics, born on 30 March 1998 as a non-profit foundation that could fund 50% of the knowledgebase. Simultaneously to SIB, the company Geneva Bioinformatics (GeneBio) S.A. was established as the exclusive commercial representative of SIB, to compensate the other 50% of the costs. GeneBio was selling licenses to commercial users, the fee depending on the number of users in the company, while academics users had free access. The royalties greatly exceeded the portion of the budget provided by the Swiss Federal Government. The Swiss-Prot group grew rapidly to the size of 80 people in 2004, while the database more than quadrupled in content in 6 years. In 2002, the funding model of Swiss-Prot changed and returned freely accessible to all the users. SIB and EMBL-EBI joined with PIR to form the UniProt consortium and applied for a NIH grant. Today the UniProt consortium has three main funders: the Swiss government (from 1996 and through SIB from 1998), recently with a 4-year grant from 2017 to 2020 accounting for about 38.7% of the total costs, an NIH grant ending in April 2018 (∼ 32.5%), funds from the European Molecular Biology Laboratory (EMBL, ∼ 25.4%) and other funding of different sources (∼ 3.4%). Swiss-Prot has also been supported by some EU funding, which ended in 2009. Now, despite the fact that about 28% of its users are from Europe, only a little portion of the curated part of the UniProtKB is currently supported by European funding.

The yearly total income of UniProt is in the order of $ 17 million (∼ €15 million), of which more than 90% is going to the staff salaries. This income hardly allows the curation of the current relevant literature⁵. However, it does not allow any expansion that is required by the fast-growing need of literature biocuration.

4 Application of the models to the UniProt case

In this section, the models presented in Section 2 are applied to the case study of the UniProt knowledgebase. For each model, the conditions to obtain an income equivalent to the UniProt annual effective costs, rounded up to €20 million, are analysed.

When possible, the analysis of the feasibility of these models is extended to a theoretical global cost of the ensemble of the bioinformatics major core data resources for life science research (i.e. repositories and knowledgebases), estimated to about €190 million. By extension and to simplify the reading, this amount will be referred to as the budget for the “total core data resources”. This value has been assessed from an estimated cost of the ELIXIR candidate core data resources (∼ €70 million⁶, per 435 million inhabitants for the ELIXIR’s Member States), extrapolated to a virtual geographical area that includes Europe, USA and Japan (respectively of 743 million, 320 million and 128 million inhabitants, for a total of 1.19 billion inhabitants). Other studies^22,23 reported different values for other groups of resources and / or infrastructure, which may be in contrast with the data presented here. Yet, it is important to emphasize that the amounts presented in this work are rough estimations and should not be taken as financial reference data: their main purpose is to illustrate how the funding models could be applied to a real case study.

All estimations are based on available data (number of users, data download, ads, etc.). The revenue values are not intended to be an exact calculation, but should be used as an indicator of the income potential for the different models. Usage data of UniProt have been obtained through Google Analytics, with adjustments for the user population as in 23,24. This triangulation leads to an estimation of unique users per month at 83,000 units.

Whenever possible, data are presented in the original currency from which they are derived, transformed in euros for ease of understanding (US$ 1 = €0.88, CHF 1 = €0.93, currency rates as at July 2017). To simplify the reading, amounts are approximated and each model is considered as a single model (i.e. no mixed model). The depth of the analysis of each model depends in general on its applicability to the UniProt case study. Some models that are theoretically applicable, but dependent on several variable parameters, are also not presented in greater detail as they would require a separate business analysis that is out of the scope of the present study. Also note that the analyses are performed under the hypothesis that the choice of the model does not influence the parameters of the same model (i.e. no feedback loop).

1. National funding. In this model, the countries having the highest access rates of the UniProt website would pay an amount to the UniProt consortium, proportional to the usage (2016 data) or to the national wealth (OECD data⁷). Four parameters have been separately taken into account for this analysis: the UniProt usage rate, the Gross Domestic Product (GDP), the Net National Income (NNI) and the R&D domestic spending. Table 2 gives an overview of the costs for the top-10 user countries, both for the UniProt case and for the total core data resources. For the first parameter, the costs respectively for UniProt and for the total core data resources are distributed among the user countries according to their usage rate. A very small percentage of the national budget (0.00025-0.00035‰) or the R&D spending (0.014‰) would allow the sustainability of UniProt. Similarly, supposing the same geographical usage distribution for total core data resources infrastructure as for UniProt, it is possible to estimate that 0.0024-0.003‰ of the total budget or 0.13‰ of the total R&D spending could sustain the total core data resources.

This model guarantees secure funds for knowledgebases, and is stable over time. It is compatible with the criteria of open access and equity for users and institutions, and coherent with the idea that the countries with the highest number of users contribute to the maintenance of the infrastructures from which they are benefitting. A contribution based on the R&D spending might be preferred to the GDP as these two do not always correlate. For a country with a large population such as India, where the expenditures in research represent only the 0.85% of the GDP (compared to 3.2% for Japan, for example), the contribution might be seen as unfairly large for the government.

At the international level, some recommendations that are consistent with this model have been recently put forward. The European Commission’s High Level Expert Group on the European Open Science Cloud (HLEG - EOSC) proposed that about 5% of the total research expenditure should be spent on properly managing and ‘stewarding’ data in an integrated fashion. The implementation of this model requires, however, that the different governments or national funding agencies agree to contribute with a fixed percentage of their R&D budgets, which in general cover other research domains other than the life sciences. This model can therefore not be put into place in a short timeframe, and governance costs may represent a considerable fraction of the funds obtained.

2. Infrastructure model. The cost of data stewardship would be covered directly by the funding agencies that fund field-related research projects. This model can be implemented as a sort of revised version of National model, model 1: funding bodies (not only governmental, but also private agencies) allocate a fixed percentage of their life science grants to a budget that is subsequently distributed to the infrastructures, knowledgebases included, according to well-defined selection criteria. To estimate the percentage needed to sustain UniProt and, more generally the total core data resources, the budgets reserved to the life sciences from five theoretically selected funding agencies, have been considered (the Swiss National Science Foundation⁸, the National Institutes of Health (NIH)⁹, the Wellcome Trust¹⁰, the Japan Science and Technology Agency¹¹, and the European Commission¹². The total yearly budget assigned to the life sciences by these five agencies adds up to ∼ €21 billion. A very small fraction of this budget, in the order of 0.1%, would then be sufficient to sustain UniProt. Only 1% of the total amount dedicated by these five funding bodies to grants in the life sciences would suffice to cover the cost of the total core data resources (0.9% of these budgets corresponds to approximately €190 million). Extending this model to other major funding agencies would increase the income and reduce the percentage needed from each agency.

This approach is very attractive in terms of equity and potential of income. It requires the major funding agencies to collaborate at an international level, and represents a significant evolution in the way how research infrastructure is funded. It also necessitates that countries that are currently not funding life science databases, or in a small proportion compared to their usage, start contributing. In general, funding agencies’ revenues can come from different sources, not only from the national budget (as in model 1); therefore, the participation of a certain funding agency to this model would likely depend on the availability of its own local budget, while the identification of the knowledgebases to which the budgets are allocated would require a selection process based on well-defined indicators (as it is currently done for grant assignments) and a lead agency or an institution that would take care of the funding distribution process. See Section 5 for an in-depth discussion about this model.

3/4. Institutional support + research project grants. These two models are equivalent to the current funding scheme of UniProt, with 63% of the budget covered by institutional support (from Switzerland through SIB and from the EU through the EMBL-EBI), and 32% by NIH funding, granted for four years until 30 April 2018.

5. Content licensing. UniProt is used by many life science companies to carry out business, research and to generate profit. The potential income of a commercial paywall is thus estimated, by assuming that all the life science for-profit companies would subscribe a licence to UniProt. A (non-exhaustive) list of the life science companies of 30 major countries in the world, irrespective of their size, was extracted¹³ together with a classification of all manufacturing companies, in terms of their size¹⁴. By assuming that the relative proportions of small, medium and large companies in the life sciences sector are similar to the proportions in the whole manufacturing area, the distribution of the life science companies in terms of their size was estimated. In this case, even low licence prices would generate an income of about €20 million, e.g.:

The licence prices were intentionally underestimated in order to balance with the overestimation of the number of subscribing companies (>15,000).

When this model was applied to Swiss-Prot in 1998, the licence fees ranged from €2,500 for small companies (typically start-ups) to €90,000 for the largest companies. At that time, the necessary annual budget for Swiss-Prot was ∼ €8 million. When calculating the revenues using these licence fees, a subscription by 1/10 of the total companies calculated above, would allow for a sustainable model for the knowledgebase. The implementation of this model requires some extra administrative costs, such as the creation of an adequate platform for the payment of the licences (probably on the order of few FTEs) and the costs for negotiation with the companies. This model has a large potential of income and allows maintaining a free access to the knowledgebase for academic users, while not for commercial users. It is though not compatible with the principles of open access and could hamper licensing and reuse of data by other resources that might see their access limited or blocked.

6. User subscription fees. As estimated, UniProt has a traffic of about 83,000 unique users/month and average monthly data download from the FTP site of 30 TB. Charging the single user with a subscription fee of €20/month would allow an income sufficient to sustain the resource. Similarly, charging the user according to data download a fee of €0,055/MB download, would cover the yearly budget of €20 million. While these amounts are comparable to many subscriptions for software or applications, this model remains inconsistent in terms of equity and open science. Moreover, the implementation of this model would also require the setup of a platform for the payments, or the adoption of an existing one, with some additional (but probably negligible) costs. As the previous model, it is not compatible with the principles of open access.

7/8. Freemium service / Razor & blades. The potential income for UniProt through these two models is difficult to estimate. Their implementation would imply that a selected part of the information (or old releases) was available for free and additional features (or the latest releases) were dependent on the payment of a fee. The data within UniProt would therefore have to be split into “free" and “not-free-but-worth-paying-for", in terms of data selection or old/new releases, which would require a more in-depth analysis and a careful selection of the type of data to charge and release. This model is currently used by some scientific journals such as the Proceedings of the National Academy of Sciences (PNAS): access to the complete PNAS Online is limited to paying subscribers and to members; without a subscription, all content older than 6 months is accessible at no cost. Similarly, TAIR adopted the same policy: up-to-date curated data are available to subscribers and one year later they become freely available for anyone to download. Interestingly, from an early analysis, the group reported that the introduction of a paywall did not decrease the use of the database¹¹. These models, however, are not compatible with the principles of open access and they also require an infrastructure comparable to the subscription model to support the fee services. Moreover, they cannot guarantee the long term survival if all the resources will have to rely on subscription fees: a paywall for accessing each resource will heavily charge the user that will inevitably choose to dismiss some of them.

9. Public-private consortium. A biotechnology/pharma company consortium financing UniProt is an option that would allow academic users a free access, with the budget of the resource being supported by a consortium of companies that make use of the knowledgebase. Yet, it is hard to estimate how many and which companies would be willing (or able) to participate, among which the cost (or part of it) would be distributed. A similar model to cover part of the costs could also in principle see field-specific companies funding the part of UniProt aligned with their interests, but without a privileged access to the data. In this way, the resource would remain open access for the users, although funded by private companies. However, the history of Swiss-Prot has shown that commercial users prefer to pay a (compulsory) licence subscription because a voluntary contribution is not easily defendable in the annual budget. The recent experience of the TAIR knowledgebase also shows that support from companies as a voluntary participation lags behind mandatory fees¹¹.

10. Advertising. This model could in principle be applied to UniProt in many different manners. One possibility is to have banner ads of related companies on the web page sides proposing pharmaceutical products, lab tools, antibodies, reagents, etc. This option has the inconvenience that the advertisements may damage the high quality image of the database, in addition to being intrusive and annoying for the users. A second possibility may be to add links to company websites that are selling products related to the proteins findable through the UniProt search tool. Also, the addition of some sponsor services, as for instance the inclusion of a comparative table of products, can be of added value to the knowledgebase. Another possibility is to collect users’ data (e-mails, searches, locations . . . ) and to exchange - provided permission from the users is obtained - the information with advertisers or partners. This is a model adopted by services as Google, Facebook and Apple, and by scientific journals such as Nature and Science. This model raises criticisms in terms of privacy and high scientific quality of the database. Moreover, the setup of an advertising platform is associated with additional costs and staff to support the structure. An advertising model requires a high volume of visitors to provide a sustainable income. To generate $ 50,000 (€44,000) per year in advertising revenues, a website needs approximately 2 million page visits per year¹⁰ The UniProt traffic of about 56 million page views per year could generate a maximum of €1.3 million, less than 1/15 of the annual budget. This model can therefore not be the unique funding source for UniProt, but rather a complementary stream to other models, as compatible with the principle of open access.

11. Wiki approach. The model does not generate a revenue, but describes a way how data could be annotated, i.e. through voluntary participation of the community. The added value of UniProt lies in the high quality of the annotated data, curated by professional expert biocurators. Public contributions cannot maintain this high level of accuracy of the data. There exist many examples of resources that adopted a “Wiki-based” approach, in genomics (GeneWiki, WikiGenes), in proteomics (WikiProteins, TOPSAN), as well as for RNA annotation (Rfam, miRBase). However, these systems still encounter many obstacles such as usability, authorship recognition, and information reliability²⁵. In addition, many of these Wiki resources are based on the integration of already processed data collected from existing knowledgebases such as UniProt. New approaches have been presented to increase reliability and usability, such as mechanisms to track authorship and to encourage community participation²⁵, but many issues remain to be addressed. For example, a Wiki approach still requires funds for the basic infrastructure (servers, technical staff, . . . ). It could therefore be implemented, combined with other more robust models, as a solution to reduce the total cost of the knowledgebase. Studies have evaluated the multiple attempts to take advantage of the significant experience of the life science community (passionate scientists, students, retired researchers, . . . ) through some sort of crowd-sourcing. Yet, crowd-sourced curation appears to have a very low participation rate. In general, the more complex a database is, the more likely professional curation is to be favoured over crowd-sourced curation^22,26. Therefore, this approach is definitely not applicable to the case of UniProt: high quality data thanks to professional expert curation are at the heart of this resource, and quality could not be guaranteed through a crowd-sourced curation.

12. Donations. Many online databases and journals rely on donations from people around the world. The most famous is Wikipedia, the free collaborative collection of knowledge. Even though Wikipedia’s content comes from active users on a voluntary base (i.e. at cost zero), the site has running costs that are covered primarily by individual donations, in addition to other funding sources that allow to sustain specific projects. In 2016, the Wikimedia Foundation received $77.2 million (€72.5 million) from 5.4 million users (∼ 1% per year of all users, with an average donation of about $15 ∼ €13)¹⁵. By applying a similar model to the UniProt case, the same fraction of users could potentially contribute with similar donations to the knowledgebase. However, donations would contribute to less than 1% of Uniprot’s budget (∼ €115,000). Also, this model is highly unpredictable, as donations depend on individuals, the awareness of the funders and a strong involvement in the cause. Moreover, some extra costs are to be included, as such a model requires setting up a fundraising infrastructure (people, campaign, department, . . . ). Yet, it is worth considering this model as a complementary model.

5 Discussion and proposal for a longterm sustainable funding model for knowledgebases

As described above, most life science knowledgebases are currently heavily dependent on grants and paid subscriptions: these funding models present many limitations that are described in Section 2. The ideal funding model for UniProt, with possible extensions to the total core data resources, should respond to the following criteria:

Table 3 summarizes the pros and cons for each model, with a focus on UniProt, and an estimation of the time frame necessary for its implementation. A complex model would obviously require a longer period of time to be put in place and accepted by the community.

A model relying on access fees (model 5, User subscription fees, as well as model 6, Content licensing) would likely guarantee the sustainability of UniProt, at least as long as the resource remains useful and has an impact for the community. If academics could have a privileged free-of-charge access, commercial entities would contribute financially to the maintenance of the knowledgebase through a subscription fee. The introduction of a paywall, even for a part of the users, would probably possible impact data reuse, access and submission. One of the principal concerns is that a paywall may prevent researchers from linking to data in other databases. And of course, scientists would need to use their grant money to pay for subscriptions. One option to recoup the usage costs could be a “virtual coins model”, in which the costs for using the resources is included directly in the grant applications and a virtual budget is assigned specifically for that purpose. In this way, the resource is maintained as long as it is sufficiently used and the researchers receive pre-paid credits to access the infrastructure. In theory, the difference with a subscription fees model is that the user doesn’t subtract part of his research budget to pay the infrastructure, as the amount that s/he needs to pay is foreseen in the project estimates. The closest scenario is perhaps the BD2K Cloud Model¹⁶, though it relates to data storage and deposition databases. It hardly applies to resources such as knowledgebases with manual curation as it might be very hard for the scientists to estimate in advance the amount of usage that they will need for a project. Moreover, this implementation is not compatible with the principles of open access. Yet, in its recent experience of the application of a subscription model, TAIR claims to minimize these drawbacks by balancing subscriber-only privileges and the publication of special releases which can be downloaded and reused by other data resources. TAIR is actually also supported by a grant from the SLOAN Foundation, which allowed the transition to the subscription-based funding model¹¹.

Another possibility is the Consortium Model, in which major biotechnology/pharma companies contribute to the budget. Currently, there exist some joint programmes, such as the Innovative Medicines Initiative (IMI¹⁷), a partnership between the European Commission and the European Federation of Pharmaceutical Industries and Associations, that could in principle also support infrastructure and data resources. However, implementing this model for more than one database (for example for all ELIXIR Core Data Resources) may hinder the negotiations with the commercial partners. Alternatively, the idea of a separate consortium for each database is likely prohibitive.

On the basis of these observations, in this work, the Infrastructure Model is proposed as a sustainable model for all the life science knowledgebases, since it is compliant with the criteria we put forward above. Its process is illustrated in Figure 2. The funding agencies distribute research grants only to research projects (but not to databases), in function of their field/topic. A percentage of each grant is retained and assigned to a budget for data stewardship and subsequently redistributed among the relevant infrastructures, including Data Management Plans providers, deposition databases and knowledge-bases.

Figure 2. The Infrastructure Model on the level of the funding agency.

On the left, the current model, in which databases compete cyclically for grants against research or resource projects. On the right the Infrastructure Model, in which the funding agencies distribute research grants only to research projects. A percentage of each grant is retained and assigned to a budget for data stewardship, and subsequently redistributed among the relevant infrastructures, including Data Management Plans providers, deposition databases and knowledgebases.

In addition, whereas UniProt’s current funding is almost entirely coming from the USA, Switzerland and the EMBL member states, this model has the advantage of distributing the cost over the countries according to the composition of the science community and thus allows a considerable diversification of the revenue streams. Importantly, at a larger scale, with less than 1 percent of the life science budget of five major funding agencies in Europe, Switzerland, Great-Britain, Japan and the USA, this model would be able to fund €190 million to cover the costs of the total core data resources. Should such a model be implemented at an even wider international level, it will involve funding agencies from other countries, thus increasing the income and further diversifying the streams. The Infrastructure model has also the advantage to scale with the amount of data that is generated. The implementation of such a model requires however the appointment of a super partes lead agency or an institution that takes care of the funding distribution process and the selection criteria. This function could be played by ELIXIR, as the European reference for the life science bioinformatics resources, or by another non-profit organization. This model could also be combined with others, such as the advertising model or some donations or institutional support.

The Infrastructure model as presented could in principle be valid in the case of a consortium of funding agencies with similar volume of grants investments. However, if there is a large discrepancy among the parties, this model turns out to be unfair, as the contribution of each agency to the total budget is directly proportional to its research spending. As a consequence, “large” funders will end up in paying the largest fraction of the figure and the model will not be fair. In this situation, a variation to the model can be conceived: funding agencies are classified by size in terms of their research spending, as “small (S)”, “medium (M)” and “large (L)” funders and contribute to the total cost with a fixed percentage, depending on their category. In this way, costs will be redistributed more evenly among the funders and spread across the entire research community. A third possibility is to setup a fixed “entry fee” from each agency, that would guarantee a minimal income. The rest of the costs are distributed among the three categories of funders, again depending on their size. Figure 3 presents a comparison of the three variations of the Infrastructure Model, with the representations of the distribution of the UniProt cost among the 5 funding agencies considered for this study (NIH, EU, Wellcome Trust, SNSF, JST, see Section 4. In Case (i), each funding agency contributes with the 0.1% of its life science budget to the total cost. As the difference in research spending of the funders is so massive (the investment of the NIH into the life sciences corresponds to more than 6500% of the investment of the SNSF), the NIH ends up in paying more than 3/4 of the total cost. In Case (ii) the five funding agencies are classified depending on their life science spending and the total cost is shared among the categories: NIH as “large”, contributing for 49% of the cost, EU as “medium”, contributing for 30% of the cost and Wellcome Trust, SNSF and JST as “small”, contributing for 7% of the cost each. As clearly visible in the figure, this model has the advantage of redistributing the costs among such different funders. Case (iii) represents the extension of Case (ii), in which a fixed 2% entry fee is required from each funder (irrespective of the size) and the rest of the cost is covered by a contribution depending on the classification (S - M - L). This last variation may be perceived as the fairest, as it allows a redistribution of the costs and ensures a minimal income. The entry fee should be then set at a level that would not discourage the small funders from participating. Worldwide, similar initiatives have already been started. At the European level, the already mentioned commission High Level Expert Group on the European Open Science Cloud (HLEG-EOSC) was created in September 2015 to provide strategic advice to the European Commission on the European Open Science Cloud initiative as part of the Digital Single Market. In its discussion on the financing of research infrastructures, including e-infrastructures (e.g. ESFRI, e-IRG and Horizon 2020-related groups), the group proposed that well-budgeted data stewardship plans should be made mandatory to all research proposals and speculated that on average about 5% of research expenditure should be spent on properly managing and stewarding data. The analysis carried out in this work demonstrated that less than half of this value could actually be sufficient to maintain the core data resources.

Figure 3. Distribution of the UniProt cost among the 5 funding agencies with the 3 variations of the Infrastructure Model.

Case (i) is the classic model, in which the cost is covered by 0.1% of life science budget of each agency. In Case (ii), the five funding agencies are classified depending on their life science spending and total cost is shared among the categories with different percentages, but constant inside each group. In Case (iii) a fixed 2% entry fee is required from each funder (irrespective of the size) and the rest of the cost is covered by a contribution depending on the classification (S - M - L), as in Case (ii).

In parallel, the USA’s NIH has launched a virtual space called Commons, a shared computing resource and a repository for data and informatics tools. In 2015, the NIH started a pilot study to test the efficacy of the Commons Cloud Credits Business Model that is designed to provide unified access to a selected choice of compute resources. In this pilot project, the researchers obtain cloud credits as part of their project grant, i.e. dollar-denominated vouchers that can be used with the cloud provider of the investigator’s choice. The cloud provider has to be Commons-compatible by meeting a set of NIH standards for capacity and capabilities. This approach is supposed to provide the researchers with a cost-effective way of accessing cloud computing resources²². However, cloud providers still rely exclusively on NIH funding.

Both these initiatives concern mainly digital research infrastructures, such as archives, storage and data stewardship, while discussions on curated databases are still in their infancy. The Human Frontiers Science Program Organisation (HFSPO) with the Global Life Science Data Resource Working Group, as well as the ELIXIR Long Term Sustainability Working Group and the OECD Global Science Forum project (GSF) are all working on the issue of sustainable business models for data repositories and curated databases. The HFSPO has recently proposed that an international coalition should be set up to support the core data resources in the life sciences. The coalition would first define indicators to establish the core data resources eligible for international support, develop models that provide free global access, and help assess the fraction (an estimation of 1.5/2% has been proposed) of total research funding for such resources^27,28. A similar project carried out by the OECD GSF is exploring the complexity of the problems connected to the future support of the data resources in the life sciences. Discussions are based on a strong consensus that core data resources for the life sciences should be supported through coordinated international efforts that better ensure long-term sustainability and appropriately align funding allowing for access at no charge.

The model presented in this work is in line with these considerations: its approach encourages equity, internationality and economic dependability, but it necessitates major changes to the way funds are distributed. It thus requires negotiating with the funding agencies at an international level, with probably less effort than with all the user country governments, and the introduction of a suitable structure to support this model in the life science community.

Data availability

All data required to reproduce the analysis presented in this study are included in the manuscript.

Notes

¹http://www.snf.ch/SiteCollectionDocuments/Dossiers/dos_OA_regelung_auf_einen_blick_e.pdf

²http://www.universitiesuk.ac.uk/policy-and-analysis/reports/Documents/2015/monitoring-the-transition-to-open-access.pdf

³https://report.nih.gov/nihdatabook/charts/Default.aspx?chartId=155&catId=2, averaged on the last 10 years

⁴Communication of the Commission ‘ICT infrastructures for e-Science’ of 5.3.2009, COM(2009) 108 final

⁵https://grants.nih.gov/grants/funding/ac_search_results.htm

⁶Corresponding to an estimated cost of the 26 candidate core data resources (5 archives, 15 knowledgebases and 6 declared as being both archive and knowledgebase, for the equivalent of 320 FTEs) submitted to ELIXIR on 1 December 2016.

⁷https://data.oecd.org

⁸http://p3.snf.ch/Default.aspx?id=AR2015

⁹https://www.report.nih.gov/award/index.cfm

¹⁰https://wellcome.ac.uk/funding/managing-grant/grant-funding-data-2015-2016

¹¹http://www.jst.go.jp/EN/JST_Brochure.pdf

¹²http://ec.europa.eu/research/horizon2020/pdf/press/fact_sheet_on_horizon2020_budget.pdf, page 4

¹³http://www.biotechgate.com/gate/v3/statistics.php

¹⁴https://data.oecd.org/entrepreneur/enterprises-by-business-size.htm#indicator-chart

¹⁵https://wikimediafoundation.org/wiki/2015-2016_Fundraising_Report

¹⁶https://commonfund.nih.gov/bd2k/cloudcredits

¹⁷https://www.imi.europa.eu/