Because of the evolution of sequencing technologies, Bioinformatics Workflow Management Systems (BWMS) are a popular software for geneticists to describe workflows for analysing genomic data. Although these systems improve development productivity, they are far from being widely accepted by this community. The lack of rigorous conceptual modelling-practices explains the complexity to adapt this genetic analysis software to context changes. In order to face this adaptation issue, we propose using the capability notion as a modelling primitive for providing a sound conceptual background. This paper analyses, from a capability-driven perspective, how daily practices in a bioinformatics SME could be represented as capabilities. From this real scenario, we state current capabilities and explain how they can be supported using current BWMS. As a lessons learned, we discuss how the introductions of capability-driven development could improve their daily work
Current DNA analyses require a complex computer procedure to perform quality
results. As some authors claim [
An interesting approach for improving this scenario is the use of bioinformatics
workflow management systems (BWMS) to describe their tailored software as
workflows made up of software components that manipulate genetic data. Current
BWMS are, to some extent, end-user oriented because they provide some
guidance for creating experiments, such as visual notations or wizards. However, when
they try to use them, BMWS lack of suitable domain abstractions [
From an Information Systems perspective, we have detected a serious problem
that affects the expected quality of those tools: the lack of sound Conceptual
Modeling practices. In previous works, we have presented a holistic Conceptual
Schema for the Human Genome (CSHG) [
In this paper we explore how conceptual modelling practices can improve the
design and development of genetic analyses. In particular, our goal is to explore
how the notion of capability can help to better understand and better design those
genetic analyses. Capability-driven development (CDD) [
The main contribution of this work is to analyse CDD in a domain such
challenging and evolving as Genomics. We want to evaluate that applying the
capability notion, functional semantics associated to BWMS are more precise, design
complexity is reduced and productivity is increased. Our first step is to identify
which capabilities geneticists demand from BWMS: following the principles of
Action Research [
The rest of the paper is structured as follows: Section 2 describes several approaches that address bioinformatics software development using BWMS. Section 3 explains the IMEGEN geneticists’ scenario for genetic disease diagnosis. Section 4 specifies a concrete set of four capabilities detected in the IMEGEN use case. Section 5 evaluates how these four different capabilities are supported by three popular BWMS. Section 6 discusses how CDD can improve the current context, and section 7 states the conclusions and the future work.
Several authors have also noticed geneticists’ difficulties when they use software
tools to accomplish their work. On the one hand, Lacroix and Menager [
Cohen, et al. [
Trying to overcome this lack of conceptual modelling-based approaches, our work contributes to genetic analysis design by: i) proposing the use of the capability notion as basic modelling unit; ii) identifying several capabilities directly associated with geneticists’ requirements; and iii) assessing how BWMS support those capabilities. Specifically, we describe four dimensions or properties that are not considered by the aforementioned approaches: the goal KPIs, the context, the capacity and the ability. With this, we introduce a modelling perspective with a more suitable abstract representation of the domain.
IMEGEN [
IMEGEN’s geneticists adopted a software solution to semi-automate this
manual procedure. But with the called Next Generation Sequencing (NGS)
technologies [
As a solution to this handicap, current BWMS aim to provide geneticists, with a platform to create their own workflows, and evolve them accordingly. The use of these environments benefits geneticists because: 1) Their interests, such as financial and temporal requirements, depend only on themselves instead of IT companies; 2) They are able to create exactly what they expect from a software product; and 3) They control the created product in case of any change is required.
However, we detected a common problem: the lack of modelling practices. For instance, the need to compare a sample sequence against a reference sequence is always the same problem, even if the evolution of technology changes continuously. This change is represented in several BWMS with the introduction of a new alignment algorithm as a new modelling primitive. Mixing the stable, conceptual part of the problem, with the more volatile, technology-oriented concrete solution, is being a serious problem shared among different BWMS.
The benefits of using BWMS and models to describe the semantics of a process are clear in this sort of situations, in which neither semantics nor the problem model has changed. Thus, evolution is described as an update model projection into concrete software technologies, such as toolkits, algorithms and information systems, which would be changed according to the situation. The contribution of this work is to use the CDD methodology to describe such processes to be implemented by a BWMS.
Using the experience gained after the study of the IMEGEN scenario, we
identified a list of mandatory capabilities to be supported in their work. According to
the metamodel for capability specification introduced in [
We use these five properties for describing and characterizing each detected capability. Additionally, for each capability an example is shown to illustrate how the capability is deployed in real practice. Next, the capabilities are described:
The genetics domain is such a young field that the best way to achieve new genetic discoveries is sharing as much genetic knowledge as possible. As a result, geneticists feed from data stored in different public repositories spread around the web. To use this data, geneticists specify queries to create datasets from several public repositories. As an example, we can consider this query: “Specify a query to retrieve a reference sequence from a gene, whose identifier is “BRCA1” from the NCBI repository” • Goal: For a giving genetic disease, retrieve all the information publicly available on the repositories list. • Goal KPI: Increase the % of detected variations with relevant information for diagnosis. • Context: Public databases are accessible using different sources: restful APIs, relational databases and HTML pages. It is common the inclusion of relevant new databases. • Capacity: a private database management system for storing indexes about previously detected variations and internal data for analysis. • Ability: knowledge about data sources provides relevant and trustful information from the genomic diagnosis point of view.
Bioinformatics community spends a lot of resources in the development of algorithms, toolkits, and web services specialized in the automation of different genetic tasks. Geneticists design their experiments by combining some of these services into a workflow. Hence, geneticists must be provided with a mechanism to design the integration and configuration of external services. Example: Integrate a custom algorithm that translates a DNA sequence into a RNA sequence. • Goal: Include into the genetic analysis external services that provide data processing functionalities. • Goal KPI: Increase the number of services currently integrated into the design environment. • Context: External services are available as web services (using SOAP or Rest interfaces) and as command-line utilities. New and updated services are published regularly according to novel research. • Capacity: Computer server for the deployment and integration of external services. • Ability: Knowledge about the genetic tasks to be performed and the functional requirements.
One of the main problems of the genetic domain is the lack of widely accepted conceptual models to express genetic data. Geneticists work with different formats that are basically raw text files following a tabular structure. It is common that new formats arise with new sequencing technologies. Hence, geneticists require a mechanism for managing new genetic data representations and extract the relevant information. Example: Use an entity “Gene” and use its attribute “Gene Name” as a parameter of a genetic task instead of column 3 of the file Gene.fasta. • • • • •
Goal: Support the management of the common data formats from sequencing technologies Goal KPI: Increase the number of supported data formats Context: Each sequencing technology could potentially provide a specific data format as there is no standard. Data is stored as flat text files that must be firstly parsed to be used by BWMS.
Capacity: Text based processing tools or scripts to transform data files into structured data, i.e inside a database.
Ability: Software programmers with the skills to implement translator modules and text parsers.
During the execution of their experiments, geneticists need to store all the data transformations and results that are obtained in every step. This information is essential to check the correct execution of the experiment and also, in order to reconstruct the experiment under the same conditions. Hence, geneticists must be able to specify the creation of result reports. Example: Specify a report that contains the date and the number of Genes obtained by a given query performed through the NCBI database.
• Goal: Create a report with the relevant data from a genetic disease diagnosis. • Goal KPI: Increase the number of results correctly described in the report. • Context: Report information changes according to the disease to be diagnosed and the customer profile. • Capacity: Reporting software for generating doc or pdf files from the results of the genetic analysis. • Ability: Software programmers with the skills to implement the report templates and the import mechanisms
From a conceptual modelling perspective, the next step is to show that the modelling of these four capabilities can improve the understanding and the subsequent implementation of genetic analysis. This modelling exercise also provides a useful framework to conceptually compare which BWMS is more suitable for deploying the capability.
Following this reasoning line, we have deployed the presented capabilities into three of the most reference open-source BWMS. Commercial tools have been discarded in this analysis, because they cannot be properly evaluated using trial or evaluation versions. We claim that the discussion reported in this section could be replicated without any significant constraint using other tools that provide a similar functionality. For the evaluation, we have used the genetic disease diagnosis process described in section 3, which was specified as a “main” capability. Using that scenario, we have configured the specific BWMS and deployed a workflow that supports each capability. Table 1 summarizes the support degree of each capability in three levels as: 1) “supported (S)”; 2) “partially supported (PS)”; or 3) “not supported (N)”. Next, the results are detailed for each tool:
C1: Access to Public Data Sources
C2: Integration with External Services C3: Conceptual description of genetic data
C4: Reporting
Taverna is an open-source environment for the design, edition and execution of
scientific workflows created by the MyGrid Team [
Taverna supports C2, because different services specifications, such as WSDL, BeanShell or Biomoby, are supported, and local tools can be integrated using a command line application. It partially supports C1 because it provides a set of services for retrieving data from different biological data sources, but it is only possible to use predefined queries. Additionally, the expressivity of the workflow language only supports the definition of specific software tools, such as “runMiraProgram” or “runBlastn2seq” instead of high-level bioinformatic tasks, such as “assembly task” or “local alignment task”.
It also partially supports C3, because conceptual descriptions of data can be
defined using the MOBY-S ontology [
Galaxy is an open-source web-based environment for the execution of biological
services created by the Galaxy Team [
Galaxy partially supports C1, as users can only retrieve data through predefined queries provided by the environment’ interfaces, for example the web search interface provided by the USCS database. Also C2 is partially supported, because there is a wide array of popular bioinformatics services provided by default. Additionally, Galaxy supports the integration of new services using programming scripts, but this approach is not suitable for geneticists that lack of programming knowledge. Regarding C4, Galaxy outputs data using a mechanism based on HTML templates. However, the usability and expressivity provided with this solution is not enough to consider this capability fully-supported.
eBioFlow is an open-source workflow management system to design and execute
biological workflows developed in the academic environment as a proof of
concept of a series of PhD dissertations [
It partially supports C1, as it provides a set of services that execute a specific query against a specific data source, but it is not possible to retrieve specific data properties. Also, C3 is partially supported, as there is a service for creating data structures according to a taxonomy of entities related with the biological and the bioinformatics domain. However, the specification of each data structure is ambiguous, relationships among structures cannot be detected and there is a confusing mixture between biological properties and computer properties. Neither C2 nor C4 are supported by this BWMS.
Using a conceptual modelling approach based on the notion of capability, we have simplified the deployment of the aforementioned process. One main improvement is that using capabilities specification, workflows are easy to understand to geneticists. They state that capabilities specification are a nice and organized documentation of their process.
Regarding the analysis, the main conclusion is that Taverna is the most complete environment, as it takes into account the aforementioned capabilities and partially supports all of them. However, specific domain knowledge and entities must be introduced to improve the end-user satisfaction. On the contrary, Galaxy provides less capabilities and its main advantage is that data retrieval and local data uploading is easier. Galaxy is specifically designed for the bioinformatics domain, it is more geneticists friendly and has a simpler workflow notation than Taverna. Its main disadvantage is that the workflow language is not as expressive as the provided by Taverna. Finally, eBioFlow provides a good workflow language in terms of expressivity and a user-friendly interface. This tool also includes some interesting features such as different workflow design perspectives, automatic creation of workflow entities from data types or partial workflow execution. But, it lacks of the advanced bioinformatics functionality that Taverna and Galaxy provide.
From this preliminary analysis, we validated that the previous issues (before applying the CDD methodology) were a consequence of modelling workflows using software-oriented concepts and the lack of documented specification. Current BWMS provide software components oriented to the solution domain, i.e, programming frameworks and command-line tools to implement specific bioinformatics functionality. In order to overcome this situation, the application of CDD is useful to improve the process understanding by end-users. Reasoning in terms of the well-supported concepts of goal, goal KPIs, context, capacity and ability, geneticists can design the expected capabilities instead of designing just workflows. This idea has been discussed with IMEGEN geneticists in order to come up with a better evolution approach. Specifically, we have found a set of improvements regarding the analysed capabilities: • Integration with public data sources: Current environments only support the execution of predefined queries against a single data source at a time. We have proposed the definition of concept models, a modelling construct proposed by CDD, specifically designed to gather all the concepts regarding the information to be retrieved. Hence, geneticists will construct queries using this conceptual model instead of programming for specific APIs or database schemas. Applying a model-to-code transformation approach is also feasible to transform a query against this model to several specific SQL queries that extract data from each biological data source. As a drawback, it will be necessary the inclusion of a mechanism that combines the query responses and expresses the resulting datasets according to the conceptual model. • Data import from genetic flat files: Some environments manage genetic data by means of text files, which can have different formats. For extracting information, users have to indicate which fields inside that file contain the data. Other BWMS provide a conceptualization of genetic data using taxonomies or biological ontologies. However, usually, these conceptualizations are ambiguous or little intuitive, and geneticists are unable to select the correct entity. We have proposed that every piece of data available in the BWMS must be managed as a domain entity. This entity, with a set of properties, represents precisely and unambiguously a concept of the biological or the bioinformatics domain. Following the approach of the previous points, an entity will be instantiated when data is retrieved in the BWMS and it will be available to become an input or output of a task of the workflow. • Reporting: Report generation is constrained to the execution of software components that generate a specific report. We have proposed the definition of a model-oriented mechanism that uses the conceptual model to create personalized reports, where all the domain data available in the environment can be selected to be reflected in the report.
While several issues have been identified in current BWMS, the main problem
is the steep learning curve. For geneticists, it is a highly time-consuming task to
fully understand and master all the features provided by a BWMS. As a solution,
we believe that geneticists should use a business environment where all the
technological and low-level details are hidden. This approach addresses the current
situation because CDD offers [
The application of CDD is not a generic approach to address any bioinformatics task, but it is useful in highly evolving processes like the diagnosis of genetic diseases. Using a CDD environment, geneticists and software analysts will be able to define their diagnosis process in an agile way as a set of bioinformatics processes and services. Additionally, process models and patterns defined as a capability can guide the generation of a workflow specification for a BWMS.
This work discusses the software issues that geneticists must face daily to accomplish their genetic analyses, and the benefits of using the CDD methodology. In order to address the unresolved features, it is discussed how CDD can be useful. The work shows how capabilities can help to put conceptual modelling in practice, in order to come up with a solution that improves current practices. Hence, we encourage the definition of capabilities specifically oriented to geneticists, whose benefits are: 1) they can focus on the experiments specification; 2) they can abstract technological details and programming issues as domain issues; 3) they can avoid the necessity to learn different BWMS specific management and features; and 4) they have flexibility to change and to evolve the processes that represent their research. As a result of this preliminary study, we state that applying a CDD approach IMEGEN genetic analyses are improved. The next steps are to go further in the specification of additional capabilities and to address the deployment of the presented capabilities.
Acknowledgments Thanks to the Instituto de Medicina Genómica (IMEGEN, http://www.imegen.es) for its support and the provision of information. This work has been partially supported by the EU-FP7 funded project no: 611351 CaaS - Capability as a Service in Digital Enterprises.