Use of Response To Terms in Annotation
An opinion by Alexander Diehl
The general definition of the "response to" terms is "A change in state or activity of a cell or an organism (in terms of movement, secretion, enzyme production, gene expression, etc.) as a result of a stimulus," where a stimulus is any conceivable stimulus a living organism might face, for example, "a stimulus from a yeast species," "a stimulus by molecules of oomycetes origin," "a misfolded protein stimulus," "an inactivity stimulus," "a stimulus indicating the organism is under stress," or "a nutrient stimulus."
This definition is incredibly broad, with no temporal limitation, and with the allowance of basically any measurable change following a stimulus as being part of the response to the stimulus. The definition provides no requirement for proof of cause and effect and no requirement for distinguishing direct responses from indirect responses.
Some annotators may thus interpret these "response to X" definitions to annotate any gene that changes in its expression level in the presence of X to "response to X." This is despite the fact that many stimuli produce global changes in mRNA or protein levels that are only indirectly connected to the stimulus through cascading cause and effect relationships. A given stimulus in a microarray experiment may affect the levels of hundreds or even thousands of genes -- should these all be annotated to "response to X" even though the sheer numbers imply lack of specificity and yield little if any information about mechanism? A separate experiment in a different cell type might yield another hundred or thousand of genes not seen in the first cell type. Pretty soon half the genome is annotated to "response to X" and what have we really learned? The ultimate result of annotating changes in expression alone to "response to" terms is a mass of annotations that provide little of value to the end user, and destroy the utility of the annotated GO terms for finding the gene products that genuinely are involved in interpreting a stimulus.
In contrast, a more limited type of "response to" definition is illustrated by the term "response to dietary excess," defined as "The physiological process by which dietary excess is sensed by the central nervous system and results in a reduction in food intake and increased energy expenditure." This definition is much more concerned with the mechanism of response, not the products of response.
This second type of definition, one concerned with the mechanism of a response, is probably a more useful approach for the GO. As things stand now, I have a reluctance to use "response to" terms for expression data, as I prefer to reserve "response to" terms for the gene products involved in the actual detection, signal transduction, gene transcription, or (occasionally) the effector mechanisms that take place in response to a particular stimulus, and not simply for the mRNA or gene product that a cell produces following a stimulus. This limits my "response to" annotations to those gene products that actively interpret a stimulus and provide a clearer interpretation of what "response to" annotations mean. However I don't know how others are using these terms.
For those who prefer the broad use of "response to " terms for annotating expression data, I can understand how in this new age where metrics are paramount that the opportunity of adding hundreds or thousands of annotations based on a single expression experiment is highly desirable. But we ought to clarify their usage in any case.
I recommend the following:
1) We need to do some kind of survey of how annotators are currently using these terms, and how different MODs want to use these terms.
2) Depending on the results of the survey, the GO consortium should decide as a whole whether we want such a broad definition for these terms, or whether a revised, more mechanistic definition is appropriate.
3) Whether or not we change the definitions, we should also do an analysis of what types of data are appropriate to annotate to the "response to X" terms, to ensure consistency in annotation practice.