This is a collection of rules for curating the biological or sample metadata of an Expression Atlas experiment, hence the rules apply to all types of experiments (microarray, sequencing, proteomics). The main purpose is to make sure curation is consistent between curators and terms can be mapped to ontologies, which helps to retrieve experiment by metadata searches and makes visualisations in Expression Atlas clean and consistent (e.g. a list of organism parts doesn't contain duplicates with slightly different terms).
These rules are basic curation style guidelines and apply to virtually all experiments.
For sample characteristics and factors, keep the values as concise as possible.
Split values in sample characteristics and factors or else it will be very hard to map the values cleanly to EFO terms.
"Characteristics [genotype] => wild type C57BL/6J" must be annotated as:
Characteristics [strain] => C57BL/6J
Characteristics [genotype] => wild type
"Characteristics [organism part] => murine bone marrow" must be annotated as:
Characteristics [organism] => Mus musculus
Characteristics [organism part] => bone marrow
"Characteristics [cell type] => thymic epithelial cell" must be annotated as:
Characteristics [organism part] => thymus
Characteristics [cell type] => epithelial cell
Lower case, use spaces (not underscores) between words, and singular.
E.g. hour not hours, embryonic stem cell not Embryonic Stem Cell.
Especially for gene names and disease names, except for terms which mean the same thing across disciplines, e.g. DNA, RNA, PBS (for phosphate buffered saline), SDS (for sodium dodecyl sulphate), EDTA (chelating agent), DMEM, RPMI (the latter two are common cell culture media).
NA for blank fields is also very problematic, since it can mean either not applicable or not available.
Here are some examples of spelling out terms in full:
Characteristics [compound] => ANIT should be 1-naphthyl isothiocyanate
Characteristics [compound] => AIR should be 5-amino-1-(5-phospho-D-ribosyl)imidazole
Characteristics [compound] => PMA should be phorbol 13-acetate 12-myristate
Other common cases of acronyms:
MEF => mouse embryonic fibroblast
TGF-beta => transforming growth factor beta
IGF-1 => insulin-like growth factor 1
TNF => tumor necrosis factor
DMSO => dimethyl sulfoxide
List is specified as atlas_property_types in perl-modules GitHub in the file
ae_atlas_controlled_vocabulary.yml.
For example, terms like sample lot number and time unit are not really attributes of samples but are technical information of samples, or simply measurement units.
If age is a range use 'to' between the values rather than a dash e.g. 3 to 5 not 3-5.
Those are examples of ambiguous annotations that you should avoid:
Characteristics [treatment] => limited
Characteristics [cell type] => resistant
Characteristics [stimulus] => stimulated
Choose which characteristics should be factors by looking at the intent of the experiment, not just what is variable. E.g. A clinical study may have included samples healthy individuals and patients from both sexes, but the researcher may not interested in sex-specific differences between the healthy and affected individuals; including samples from both sexes may be simply for eliminating sex-specific bias in data analysis.
The allowed values should align with the values recognised by the atlas configuration generation script as "reference" (listed in atlas-prod config file mapped_reference_assay_group_factor_values.xml). For empty terms it is good to have consistency e.g. compound - none; stimulus - none; infect - none.
| Attribute | Values to use |
|---|---|
| genotype | wild type genotype |
| phenotype | wild type |
| disease | normal |
| compound / irradiate | none |
| transfection | mock or control or vehicle |
| RNA interference | scrambled siRNA |
| infection | sham or none |
| diet (for mouse) | chow |
| immunoprecipitate | input DNA |
| stimulus | none |
| immunophenotype | live cell or unsorted |
Some experiment designs require special rules to follow to make annotations consistent.
E.g. cells carry a construct that causes repression of gene expression when activated by doxycycline.
Use Characteristics[genotype] (and Factor Value[genotype] if variable) to annotate with info about the construct.
Use Characteristics[phenotype] (and Factor Value[phenotype] if variable) to annotate with info about the
gene expression/repression caused by compound. For example:
| Characteristics[genotype] | Characteristics[phenotype] | ... | FactorValue[phenotype] |
|---|---|---|---|
| FOXP3-tet-off | expression of FOXP3 | ... | expression of FOXP3 |
| FOXP3-tet-off | doxycycline-mediated repression of FOXP3 | ... | doxycycline-mediated repression of FOXP3 |
We use factor values compound and dose for any chemical. Stimulus is the preferred term for peptide and protein
stimulations (e.g. Interleukin-2). Compound should always be accompanied by the dose and its unit.
The compound should be found in CHEBI ontology and the unit must be in EFO.
| Factor Value[compound] | Factor Value[dose] | Unit[concentration unit] |
|---|---|---|
| phorbol 13-acetate 12-myristate | 10 | millimolar |
If we have multiple compounds as a single factor, we can combine them into a single value under "compound" (and include the respective dose), for example bovine serum albumin; 0.06 millimolar and sodium hydroxide; 0.4 millimolar. The same applies for two doses of the same compound: methotrexate; first dose 120 milligram per kilogram; second dose 60 milligram per kilogram. Leave the dose factor blank in such a case.
If a study involves both normal and tumor samples from the same patient, this should be coded as:
| Characteristics[disease] | Characteristics[individual] | Characteristics[sampling site] |
|---|---|---|
| breast cancer | patient 1 | normal tissue adjacent to neoplasm |
| breast cancer | patient 1 | neoplasm |
| normal | healthy_guy_X | normal tissue |
Characteristic[Individual] is used to show which samples come from the same patient.
Disease refers to the disease of the donor and will be cancer even for the sample from the healthy tissue.
Generally use the main ontology label, not the synonyms and the most specific term, e.g. heart ventricle instead of heart.
If the term is not in any ontology, look up the previous zooma mapping.