The EXACT description of biomedical protocols.

MOTIVATION: Many published manuscripts contain experiment protocols which are poorly described or deficient in information. This means that the published results are very hard or impossible to repeat. This problem is being made worse by the increasing complexity of high-throughput/automated methods. There is therefore a growing need to represent experiment protocols in an efficient and unambiguous way.
RESULTS: We have developed the Experiment ACTions (EXACT) ontology as the basis of a method of representing biological laboratory protocols. We provide example protocols that have been formalized using EXACT, and demonstrate the advantages and opportunities created by using this formalization. We argue that the use of EXACT will result in the publication of protocols with increased clarity and usefulness to the scientific community. AVAILABILITY: The ontology, examples and code can be downloaded from http://www.aber.ac.uk/compsci/Research/bio/dss/EXACT/.

1 INTRODUCTION

‘Everything is vague to a degree you do not realize till you have tried to make it precise.’

Bertrand Russell

The ability to repeat a published experiment protocol is the foundation stone of laboratory science. It is widely accepted that for new knowledge to be published in a scientific journal the protocols used to derive that new knowledge must also be published. This is essential to validate that the process by which the knowledge was inferred was not flawed in any fundamental way, and to ensure that the result was not caused by some chance event. In order to repeat a protocol it must necessarily be described in sufficient and unambiguous detail to enable another agent (human or machine) to be able to replicate the original experiment actions. With the increasing complexity of experiment methods, the description of laboratory protocols is becoming correspondingly more complicated and intricate. This means that there is a growing technological need to be able to represent experiment protocols in an efficient and unambiguous way.

We propose the Experiment ACTions (EXACT) ontology as the basis of a method of representing biological laboratory protocols. EXACT provides a model for the description of experiment actions and it can be used for the fully formalized representation of protocols. It can also be combined with other formalisms for the description of bio-medical investigations.

The rest of this article is organized as follows: Section 2 provides the background to this work, Section 3 provides a detailed description of the proposed ontology of experiment actions and Section 4 demonstrates the application of the ontology to the formalized description of two protocols: for creating competent cells and for compound library replication. In Section 5 we describe the opportunities for the application/implementation of EXACT and finally Section 6 provides discussion and conclusion.

2 BACKGROUND

2.1 Current problems

The degree of information granularity present in many published protocols is often insufficient to allow the method to be repeated successfully. An optimization period is then necessary to bridge the gap in knowledge between the published protocol and one which works reliably. Knowledge of how best to implement an existing method is regarded as a group's intellectual property and is often not included in published manuscripts. Each and every time research results are published with insufficient information in the materials and methods section this duplication of labour is repeated, adding to inconvenience and cost.

A manuscript published by Akada et al. 2006 illustrates the difficulty in repeating another researcher's protocol when not all the necessary information is provided. The manuscript outlines a novel protocol for gene deletion in Saccharomyces cerevisiae. The protocol focusses on the generation of a deletion cassette through the fusion of two DNA fragments. The manuscript provides ample information on strains, media, primer sequence and polymerase chain reaction (PCR) conditions. However, when repeated in our laboratory the deletion cassette could not be generated. Personal communication with the author revealed that the two DNA fragments require gel purification before a successful PCR fusion can occur. This proved to be a vital step yet was not included in the published manuscript.

The excerpt shown in Table 1 describes the first stages of making yeast cells competent and was taken from the High Efficiency Transformation of Yeast protocol published in Methods in Yeast Genetics Amberg et al., 2005, a text book routinely cited in published papers.

Table 1.

High-efficiency transformation of yeast, methods in yeast genetics

1. Inoculate 4 ml of liquid YPAD or 10 ml of SC and incubate with shaking overnight at 30○C

2. Count overnight culture and inoculate 50 ml of YPAD to a cell density of 5 × 106/ml culture

3. …

The protocol is summarized in point form using natural language. This can lead to ambiguous statements with unclear objectives. Point 1: does the word inoculate mean using a single yeast colony from a solid media plate or can the liquid YPAD (Yeast extract, Peptone, Adenine hemisulfate, Dextrose) be inoculated from another previously inoculated YPAD liquid culture? The statement incubate with shaking, is this at 20 rpm or 400 rpm? Does overnight mean 12 or 24 h? Point 2: it states count overnight culture, which only suggests that the person executing the protocol needs to calculate approximately how many cells are present in the overnight culture. How this estimate is achieved is not stated. Having statements which can be interpreted in different ways introduce inconsistencies in how the protocol is executed. This can introduce noise and contribute to inaccurate findings. In this instance the author has nothing to gain from omitting information from the method. The target audience of this protocol are yeast biologists and therefore the author assumes that the reader has a degree of prior knowledge. However, this is not always the case. A researcher new to this field may mis-interpret a statement resulting in an inaccurate objective. It is also possible to envisage information being deliberately left out of published protocols, particularly if omitted information reduces the impact of the findings. This information could suggest that the findings cannot be reliably reproduced or could reflect poorly on the success rate of a novel technique.

2.2 Existing approaches

The requirement for an efficient representation of experiment protocols is recognized as a pressing problem, and several other projects are applying ontologies to the formalization of knowledge about experiment data. The Microarray Gene Expression Data (MGED) ontology is one of the pioneering attempts to use an ontology to record information about experiment data (Whetzel et al., 2006b). The MGED ontology was designed to formalize the descriptors required by the Minimum Information About a Microarray Experiment (MIAME) standard for capturing core information about microarray experiments (Brazma et al., 2001). Many journals (∼50 thus far) require {MIAME} compliant data as a condition for publishing microarray-based papers. This is a trend that looks set to continue for other accepted ontological standards. The Minimum Information About Proteomics Experiment (MIAPE) standard supports proteomic experiments (Taylor et al., 2007). The Metabolomics Standards Initiative (MSI) ontology working group is building an ontology to facilitate the consistent annotation of metabolomic experiments (Sansone et al., 2007).

Minimum Information for Biological and Biomedical Investigation (MIBBI) is a web-based resource designed to act as a one-stop-shop for those seeking for or looking to contribute to Minimum Information (MI) checklists. According to MIBBI's website these checklists are intended to promote transparency in experiment reporting, enhance accessibility to data and support effective-quality assessment, thereby increasing the value of a body of work. The MIBBI project maintains a web-based resource for extant checklist projects, complementary data formats, tools, controlled vocabulary and databases. MIBBI aims to provide guidelines for checklist development, both by increasing connectivity between MI checklist development projects, and by disseminating best practise both in relation to process (such as open mechanisms to receive and respond to public comment) and presentation (e.g. use of shared language, documentation style and structure, production of user-friendly summaries). Checklists developed using MIBBI guidelines focus largely on capturing the minimum information needed to usefully annotate data generated by biological/biomedical investigations. Information pertaining to wet-lab processes such as experiment execution(s) is described using natural language with the aid of some controlled vocabulary. However, the degree of information granularity in these descriptions is thus far at the authors discretion. These descriptions may be sufficient to make effective use of reported data but are likely to be insufficient to independently repeat the experiment actions used to generate the date.

PRoteomics IDEntification (PRIDE) is a data repository, supported by a combination of tools, standards and infrastructure for the description of proteomic data (Martens et al., 2005). PRIDE's schema presents a minimum of information about protein identifications. PRIDE's top level structure contains the part annotated with key words used to mark the type of method used to generate the proteomic data.

MGED, MIAME, MIAPE and PRIDE are ontologies primarily focused on developing controlled vocabulary and descriptors for high-throughput strategies such as mass spectroscopy and array-based comparative binding assays. These ontologies are centred around the annotation of data. Protocol information is only present at a level of detail, which is sufficient to describe the data. None of these projects provides a detailed enough formalism for the representation of experiment actions.

The Functional Genomics Investigation Ontology (FuGO) project Whetzel et al., 2006a), and its successor the Ontology for Biomedical Investigations (OBI) project, are developing an integrated ontology for the description of biological and medical experiments and investigations. This ontology aims to model the design of an investigation, including the protocols, instrumentation, materials used and the data generated. OBI has not yet been released, but it already has the key classes for the description of protocols: , , . The generic ontology of scientific experiments (EXPO) aims to formalize domain-independent knowledge about the organization, execution and analysis of scientific experiments (Soldatova and King 2006). This ontology has the class and defines some of its properties: has Goal, has Object, has Instrument, but there are no subclasses specified. Our proposal differs from the existing ontology-based approaches for the description of experiment protocols by suggesting a meta-language for the description of experiment actions and their properties. EXACT provides a formalized representation of the domain that is not sufficiently covered by any other ontology.

In theoretical computer science, process algebras have been used to specify and reason about descriptions of processes and actions. Process algebras are algebraic systems for the manipulation of elements of processes (the individual elements being actions or events). They define laws governing the sequencing, composition and synchronization of actions. Leading examples of process algebras are the Communicating Sequential Processes (CSP), the Calculus of Communicating Systems (CCS) and the Algebra of Communicating Processes (ACP) (Bergstra and Klop 1984; Hoare 1985; Milner 1980). For example, a process algebra would provide laws stating that: is equivalent to the choice of Process algebras in biology have generally been used as modelling languages for biological systems rather than as a way to specify experiment actions. The ontology we propose provides much more detail than process algebras are usually designed to give, but could be used together with a suitable process algebra for verification and other algebraic reasoning over protocols. The process algebra for biological protocols would need to represent parameterized actions (e.g. to incubate at 30○C). It would also require the ability to represent state changes (e.g. CSP∥B, Treharne and Schneider 2002, which combines the CSP representation of processes with the B formal language to represent changes in state).

Logics for agency have also considered some of the issues that we deal with in this work. In particular agents are described by their actions and goals (desires/intentions). There are many logics for agency, each allowing different expressiveness and covering areas such as belief, knowledge, possibility, time, branching, the relationships between actions and goals and the distinction between understanding what must be done and why it must be done.

In EXACT we first provide a vocabulary and ontology, and then begin to look at the grammatical aspects of describing experiment actions.

2.3 Our proposed solution

To develop EXACT, we first analysed protocols from several bio-medical domains, including functional genomics, metabolomics and drug screening, as well as protocols published in Nature Protocols. We then consulted with biologists, microbiologists, biochemists and chemists with experience in the execution of these protocols to clarify ambiguous statements and to enrich the protocols with as much information as possible. This helped to capture the precise meaning of each experiment action performed. General concepts were abstracted from these experiments actions and were used to develop the ontological classes. The scientific experts then used these classes to try and represent their own protocols. After many painstaking rounds of consultation, classes were added and removed or changed in the ontology to help better represent the actions performed in various protocols. The EXACT hierarchy of experiments is sufficient to formalize many of the protocols used in our labs. However, as we formalize more and more protocols using EXACT, its class structure will grow and evolve to meet the needs of new methods and techniques.

3 AN ONTOLOGY OF EXACT

An ontology of EXACT aims to provide a structured vocabulary of concepts for the description of protocols in bio-medical domains. Our ontology intends to be compatible with other formalisms, to share and reuse already formalized knowledge. For example it reuses classes from the phenotypic qualities ontology PATO, OBI and the W3C Time Ontology (OWL-Time). EXACT is expressed in OWL–DL and was developed using the Protégé ontology editor.

The main part of EXACT is a hierarchy of experiment actions. This hierarchy was created using a classification based on goals of actions. The experiment actions are divided into three groups according to their goals: In defining these groups of actions we follow the classes of elementary processes used by Noy (1997). Our approach differs by separating ‘what is done’ from ‘how it is done’. The same goal can be achieved by many different actions. For example, the goal may be achieved in various ways: by the experiment action , by the experiment action or by other actions.

separation;

transformation;

combination.

Experiment actions that provide a mode of transformation are classified into the following subclasses: Figure 1 shows our classification of experiment actions according to their goals.

Fig. 1

EXACT classification of experiment actions.

a mode of property transformation, with such experiment actions as , , ;

a mode of transformation of spatial location, for example ;

a mode of transformation of time, for example ;

a mode of category transformation, with such experiment actions as , , .

Many experiment procedures have experiment actions that are executed only if a certain condition is valid. For example in our experiments with yeast, the optical density (OD) of yeast cells should be between 0.6 and 1.0 before processing. If the OD is too low, the culture must be incubated for longer. In order to represent conditions, EXACT defines the class with the sub-classes , , and . Each condition has a ‘boolean expression’, a ‘yes-command’ for execution if the value of the expression is true and a ‘no-command’ for execution if the value of the expression is false. Pre-conditions and post-conditions can be used to check whether materials and instruments are ready for execution of experiment actions, whether final volumes of solutions are correct, or whether objects are in the correct locations. Such checks during running of experiments are important to prevent errors. Store conditions are used to indicate when it is possible to temporarily stop execution of the protocol and put materials in a store under the defined storage requirements.

EXACT also defines a set of command actions, which control the flow of execution of the protocol: , , , and . is the null action that does nothing at all (but is useful as the yes-command for pre- and post-conditions, and the no-command for store-conditions). terminates the flow of execution immediately and is used as the no-command for pre- and post-conditions. is used in conjunction with all four conditions described above to test the expression and execute the yes-command or no-command as appropriate. is the action of storage and is used as the yes-command of the Store-condition. is a command that moves the flow of execution to an action elsewhere in the protocol. The command actions currently defined by EXACT are minimal and are likely to be enhanced in the future by more complex constructs such as loops.

The class is used in the ontology to describe that some entities can play a certain role. A location can be a start or end location, a piece of equipment can be a start or end container; a location can be a lid location, etc.

Apart from experiment actions that are performed by manipulating dependent variables of an experiment, EXACT defines the class with instances that have the goal of preparing for experiment actions. These actions are considered as preliminary to later actions. EXACT also defines the class with instances for recording measurements and observations that have the goal of preservation of information. EXACT includes a hierarchy of instructions with the classes e.g. and e.g. (taken from Nature Protocols) that can be ignored by automated agents executing protocols, but warn human users to take extra care.

EXACT is available in two versions: EXACT/EXPO is compliant with EXPO (Soldatova and King 2006) and more suitable for automated laboratories; EXACT/OBI is more suitable for using within OBO communities. EXACT/OBI provides an explicit mapping to OBI (the current draft March, 2008).

The principal difference between these two versions is in philosophical foundations. OBO ontologies are based on a philosophy of reality and do not include abstract entities. This does not put considerable restrictions on the description of the existing protocols as most protocols are designed for execution of experiments in the real physical world by manipulating real physical objects. The results of our research show (King et al., 2004; Soldatova et al., 2006; Whelan and King 2008) that the representation of logical and mathematical objects (i.e. sets, relations, facts) and other entities within a computer system as abstract entities provides a clearer description of computational experiments and experiments executed in automated laboratories.

Philosophers have argued about fundamental ontological questions for at least two and a half thousand years, and we do not wish to enter these debates. What we need to do is to make practical decisions about how best to describe protocols. We believe that supplying different versions of EXACT is the best way to deal with conflicting upper ontologies. Hopefully the two versions of EXACT can be merged when a philosophical solution is found that is suitable for all needs.

EXACT/EXPO has only two abstract entities and , which are defined as the value of a statement that corresponds/does not correspond to reality. These classes are used in pre- and post-conditions of actions. EXACT/EXPO is designed to be compliant with an ontology for automated laboratories, which we are developing at Aberystwyth, UK. In EXACT/OBI these classes are defined as subclasses of the class , which was recently introduced into OBI (January, 2008). The class is a subclass of the class . Table 2 shows an explicit mapping between the top classes of the two EXACT versions.

Table 2.

EXACT/EXPO – EXACT/OBI mapping of the top classes

EXACT/EXPO	EXACT/OBI
<process>	<BFO: occurent>
<object>	<BFO: continuant>
<proposition>	<OBI: information entity>
<quality>	<BFO: quality>
<role>	<BFO: role>
<abstract entity>	<OBI: information entity>

EXACT/OBI defines a mapping of the EXACT/EXPO top classes to the leaf classes of Basic Formal Ontology (BFO) (Grenon and Smith 2004) without considerable loss of semantics and can be reused within OBO ontologies.

EXACT aims to follow OBO Foundry principles: ‘the ontology is open and available to use by all’, ‘is in a common formal language’, ‘includes textual definitions of all terms’, ‘uses relations which are unambiguously defined’, it is orthogonal to OBO ontologies and it follows the naming convention of (Schober et al., 2007).

The current version of EXACT does not yet include axioms. We are collecting statements about experiment actions and plan to include them in the form of axioms in the next version. Here are some examples of such statements: The second statement tells us that in order to mix components, the components must be moved to the same location.

Apart from the well-defined foundational relations is_a and part_of, EXACT includes the relations from the OBO Relational Ontology (RO) (Smith et al., 2005) located_in, has_participant and has_agent, the relations has_role and has_quality that are used in OBI, DOLCE (Gangemi et al., 2003) and HOZO (Kozaki et al., 2002), and a relation has_proposition (or has_information for the EXACT/OBI version).

The specialization of BFO in representing real world entities is reflected in the set of RO relations. RO relations are not suitable for linking physical entities to information entities. The set of RO does not allow to easily represent such knowledge as ‘an experiment action has a goal’, ‘an action is conditional’, ‘an investigator has a plan’. EXACT includes the relation has_proposition to fill this gap. The relation allows the connection of an agent of a process with a certain portion of information that is essential for participating in the process. We define this relation following the methodology suggested in Smith et al. (2005). First, we add one more relation to the ‘pain of infinite regress’ of primitive instance-level relations: There is a primitive relation between an agent of a process, a proposition and a time, where a is an agent, p is a process, i is a proposition and t is time.

Second, we define a class-level relation using this primitive instance-level relation: where A and I are classes of agents and propositions. We can express ‘an experiment action has a goal’ with this relation as follows: an agent of an has_proposition .

EXACT is a modular ontology. It defines a conceptual scheme for describing experiment actions and their properties. To represent individual experiment actions it is necessary to import individuals of the classes