Morphological variation and strain identification of insects using wings and I3S.
Novel insect identification techniques often lead to speculation on whether the method could cope with any intraspecific variation that might occur in a species. Using I3S Classic (Interactive Individual Identification System, Classic) and images of mosquito wings, different mosquito strains were tested with a copy of the strain present or absent from the database which contained images of other strains of the test species. When a wing image of the exact species, strain and sex was present in the database, there was 100% (or near 100%) retrieval of the correct species and strain at rank one. When the exact strain was absent from the database, but other strains of the same species were present, the retrieval rates at rank one were again high (100%) in the majority of cases and when they were not, the correct species was generally retrieved at rank two. Out of 40 different species and strains tested, only three were significantly different at rank one when the exact strain was absent from the database. In general, images of field strains selected for each other and therefore were similar to each other in greater numbers and instances than for the laboratory strains tested. When a copy of a strain was absent from the database, but other strains/sibling species were present, I3S retrieved the correct strains/sibling species at rank one in the majority of cases. In the one case of transgenic mosquitoes tested, I3S could reliably be used to identify transgenic mosquitoes from the parent stock as they were retrieved 100% at rank one when both the transgenic and unmodified parent strains were present in the database. This indicates the potential of using I3S to distinguish transgenic or other selectively bred strains from a parent strain, also selectively bred and wild mosquitoes, at least in the first phase after field release. Similarly, hybrid strains, sibling species and members of species complexes as in the Anopheles gambiae species complex could also be correctly identified when copies of all the relevant species/strains/siblings were in the database. This contradicts the belief that only molecular characterisation could separate A. gambiae s.s. from A. coluzzii, or A. arabiensis; I3S could accurately separate them all. I3S worked as it was set up to do, retrieving closely resembling images of the test insects from the database and ranking them in order of similarity. Dealing with any intraspecific variation was therefore not an issue if the software (I3S) was used systematically. I3S complements molecular and traditional taxonomic methods for species identification and the separation of sibling complexes and strains. In future, it should become the norm to maintain databases of insect wings and other body part images for use in image recognition.