Fish Parasite Ecology Software Tools (Fish PEST) is a web tool aimed to model the distribution of parasites on fish hosts. Fish PEST integrates parasitological data with ecological, biogeographic and phylogenetic information of fish hosts. Although originally developed as a standalone project, Fish PEST has been deeply integrated in Fishbase, from which it takes all data about fish ecology, biogeography and phylogenesis. Fish PEST allows users to:

   - Create a model host range for a parasite with PaNic (Parasite Niche Modeler);

   - Create a model host range for a parasite from a custom host range with PaNic custom section;

   - Create a model parasite list for a host species with PaCo (Parasite Co-occurrence Modeler);

   - Create a host/parasite list or matrix from the internal database with PaL (Parasite List Generator).


If you use FishPEST in your analyses, please cite:

Strona G. and Lafferty K.D. (2012). FishPEST: an innovative software suite for fish parasitologists. Trends in Parasitology 28: 123.


If you use PaNic, please also cite:

Strona G. and Lafferty K.D. (2012). How to catch a parasite: Parasite Niche Modeler (PaNic) meets Fishbase. Ecography 35: 481-486.


If you use PaCo, please also cite:

Strona G. and Lafferty K.D. (2012). Predicting what helminth parasites a fish species should have using Parasite Co-Occurrence Modeler (PaCo). Journal of Parasitology (in press). doi: http://dx.doi.org/10.1645/GE-3147.1.

INDEX OF CONTENTS:

1 - Data


2 - PaNic
    2.1 - Quickstart
    2.2 - Results Page
    2.3 - Technical Details
        2.3.1 - Parasite Niche Calculation
        2.3.2 - Outliers
        2.3.3 - Phylogenetic Constraint
        2.3.4 - Biogeographic Constraint

3 - PaCo
    3.1 - Quickstart
    3.2 - Results Page
    3.3 - Technical Details
        3.3.1 - Algorithm
        3.3.2 - Optimal Model Search

4 - PaL
    4.1 - Create Lists and Matrices
    4.2 - Search DB
    4.3 - Upload Data

5 - Troubleshooting

6 - References

7 - Database Sources

8 - PaNic Tutorial

Fish PEST relies on an internal database including more than 16000 validated host-parasite records for Acanthocephala, Cestoda, Monogenea, Nematoda and Trematoda. The host-parasite records derive from a much bigger list containing more than 64,000 host parasite records, which was collected from scientific literature, Internet databases and museum collections, and which was reduced to a quarter during the validation process. Host scientific names were validated according to Fishbase (Froese and Pauly 2011), while parasite scientific names were validated according to the Catalogue of Life (Bisby et al. 2011) and WoRMS (Appeltans et al. 2011). Invalid synonyms were replaced with the corresponding current valid names, with the exclusion of any ambiguous record. Host or parasite records at a level higher than species were excluded as well. Periodical validation of fish and parasite data is scheduled on a bimestral basis (according to Fishbase standards). Considering the purposes of FishPEST, researchers are encouraged to submit their peer reviewed data (see Upload Data section). Fishbase Species Ecology Matrices were used to compile an eco-biological matrix for more than 27,400 fish species. Of the parameters available in Fishbase, PaNic uses maximum length, growth rate (rate at which the asymptotic length is approached, termed K in Fishbase), life span, age at first maturity, and trophic level (please refer to Fishbase documentation for specific details). For each species, habitat and geographical information available from Fishbase were included as well.

Parasite Niche Modeller (PaNic) is a web-tool for fish-parasite niche modeling. PaNic uses eco-biological parameters of the known hosts of a parasite species or genus as inputs to the Bioclim algorithm (Nix 1986), which computes niche boundaries for a list of hosts. All candidate fish species are then evaluated for their overlap (compatibility) with known hosts of the parasite. In order to build a model, PaNic requires the following steps (see scheme):
1) You create a model host set (C) by selecting a parasite species/genus (A), or by custom host selection (B). This tells PaNic the types of hosts your parasite is known to use.
2) To get a smaller output of potential hosts, you can filter the input list by imposing phylogenetic and biogeographic constraints on the model hosts (D).
3) From the parameters of the filtered list of known hosts (E), PaNic creates a niche model for the parasite (F).
4)PaNic then applies the parasite niche model to a projection host set (G), selecting members of the projection set with eco-biological parameters closest to niche model (H), which are then passed through the user-selected biogeographic and phylogenetic filters to return a list of the most compatible hosts (I). For further details on PaNic, please refer to Strona and Lafferty (2012a).

1 - a) Select a parasite species or genus. You can select one from the drop-down menus or, alternatively, you can search for a particular species by inserting its scientific name in the input text areas. The first character of the Genus must be a capital letter. If the inserted scientific name is misspelled or does not match any of the species available from the internal database, PaNic will return an error page (if the latter, try a custom model). After parasite selection, PaNic populates a known host set, or KHS, which is the set of known fish hosts for the selected parasite.

1 - b) Selecting a parasite species from the internal database is not mandatory to build a model. Custom model section allows you to create models for parasite species not included in the list (or for which you have additional data on host use). To build a custom model, manually select a set of known host species (ideally >10).

2) Select the eco-biological parameters (maximum length, growth rate, life span, age at first maturity, and trophic level) you want to include in the model. You can also attribute different weights (W) to each parameter. By default, all the parameters are included with equal W (9). See Fishbase documentation for specific details on eco-biological parameters.

3) Create the projection host set (PHS), i.e. the set of potential hosts you want to test for compatibility with the selected parasite. You can simultaneously select projection hosts by family, locality or as individual species. You can also decide to include in the projection set all the available host species. If you choose this last option, the list of less likely compatible hosts and the graphics will not be displayed in the results page, in order to save time and server memory.

4) Apply one or more filters to the projection set. You can decide to include in the projection set only hosts occurring in marine and/or brackish and/or freshwater environment. You can also choose to apply a phylogenetic or a biogeographic filter. Increasing the value in 'Constrain Host Phylogeny' field, you will make sure you are including in the PHS only host species phylogenetically similar to those of the KHS. Similarly, by increasing the value in 'Constrain Host Geography' you will make sure you are including in the PHS only host species with a geographical distribution similar to that of the members of KHS.

5) Tune the main model parameters:
    - Modify the value of Niche Restriction in order to increase or decrease the power of outlier detection.
    - Cutoff is a coefficient of the algorithm PaNic uses to compute niche models. Increase its value if you want to make wider niche intervals for the selected parasite (decrease for tighter intervals).
    - Compatibility threshold expresses the minimum percentage of fitting between niche intervals for the selected parasite and ecological parameters of a member of the projection host set, for the latter to be considered likely or less likely compatible. Increase this value if you want to create more restrictive models.

6) You can choose if you want the results page to display a list of less likely compatible hosts and/or graphics that indicate the niche visually.

7) Submit your query.

The results page is organized in two parts. The first part summarizes the main features of the model, comprising:

    - the host species for the selected parasite;
    - any known hosts that appear to be outliers of the set of known hosts;
    - the overall phylogenetic and biogeographic similarities of the known hosts;
    - eco-biological variables excluded from the model;
    - relative variation of the eco-biological variables, computed for each variable (rVEV) as the ratio between its standard deviation in the KHS and its standard deviation in the PHS;
    - the lists of hypothesized likely and less likely compatible hosts.

You can save a full summary of the model to a text file by clicking on the 'save' button.

The second part of the results page (that will not be displayed unless the user has checked the graphics options in the main page) contains, for each proposed host species, a graphical representation of the relative value of its ecological parameters with respect to the computed parasite niche.

Niche boundaries (NB) are calculated according to the Bioclim algorithm (Nix 1986). In most niche models, spatial correlations with environmental variables (e.g., temperature, rainfall) from known locations can be used to extrapolate broader scale distributions (Gusan and Thuiller 2005). PaNic modifies this technique to focus on host characteristics. For each host eco-biological parameter (E), lower and upper NB are respectively calculated as:

    mean(E) +/- (StDev(E) * a*10^(-1)),

where a is a cutoff value that is set, by default, to 7 (which is an approximation of Bioclim standard value, see Nix 1986).The higher the value of a, the wider the resulting niche intervals. For each parameter, besides NB, compatibility boundaries (CB) are calculated as well. Lower and upper CB are calculated as the minimum and maximum of E values among KHS. For each member of the PHS, a high compatibility score (hc) is computed as:

    

where:

- Wi weight assigned to i-th E;
- n= number of Es included in the model.
- Xi = binary operator indicating whether or not i-th E fits into NB.

Similarly, a compatibility score (c) is computed with an expression that is the same as the above, with the only difference that Xi will indicate whether or not i-th E fits into CB. A member of PHS is considered likely compatible with the computed niche model if:

    hc > C,

while it is considered less likely compatible if:

    hc < C < c.

After a KHS has been defined (both from the internal database, or through custom host selection), PaNic performs a procedure to detect outlier hosts that will be removed from the definition of the niche. Outlier hosts are not incorrect host/parasite records, so much as hosts in the KHS that differ significantly from the others in their eco-biological parameters. Hosts are outliers if their average Mahalanobis distance from each other member of the KHS (computed on the basis of the considered eco-biological parameters) is larger than n times the overall average Mahalanobis distance among each member of the set, where n is equal to:
(1+(10-NR)*0.1). NR is set to 0 by default (minimum restrictiveness), but users are allowed to select different values. The higher the value of NR, the lower n will be, leading to fewer hosts included in the model.

The Phylogenetic Effect Value (PEV) weights the importance of coevolutionary processes which can lead to host specificity. Parasites vary in host specificity and users might want to consider the tendency for coevolution when applying niche information. In PaNic, the potential effect of coevolution can be accounted for with a phylogenetic proximity index (PPI) that is calculated for each species of the PHS as:
PPI=(s+ g+ f)/3N,
where s, g and f are respectively the total numbers of species, genera and families it shares with the members of KHS, and N is the size of KHS. The value of s will be echoed in g and f, and g will be echoed in f (but not vice-versa), providing a balance among different taxonomical levels. The greater the overlap between the phylogeny of the KHS and the potential host species, the closer PPI is to 1. If there is no overlap even at the family level, PPI = 0.
The weight of a phylogenetic constraint can be included by increasing the value of PEV from 0 (default, no effect) to 9. A member of PHS will be considered compatible to the selected parasite only when PPI > (PEV+1)*0.05. To give an indication of host specificity within the KHS, PaNic calculates a measure of phylogenetic proximity (PP) or:
(S + G + F)/3N,
where S, G and F are, respectively, the number of shared species, genera and families among the members of KHS, and N is the size of KHS. PP is maximum (= 1) when all members of KHS are subspecies of the same species. It will be 0 when each member belongs to a different family. High values of PP (>0.5) should encourage users to account for coevolutionary effect in their model, by increasing PEV.

The Biogeographic Effect Value (BEV) weights the importance of geographic limitations on parasite distributions. For instance, some parasites are cosmopolitan, while others are restricted to oceans or latitudinal ranges, islands, or continental regions Locality records for the fish of KHS are compared to those of each member of PHS. For each member of KHS a biogeographic Compatibility Index (BCI) is calculated as:
sl/L,
where sl is the number of locality occurrences it shares with the members of KHS, and L is the number of localities where at least one member of the KHS occurs. The weight of a biogeographic constraint can be included by increasing the value of PEV from 0 (default, no effect) to 9. A member of PHS will be considered likely or less likely compatible to the selected parasite only if BCI > BEV*0.1.
An approximate indication of biogeographic range (BR) within the KHS, is calculated as:
L/Lt,
where Lt is the total number of locality records for KHS, and L is the number of localities where at least one member of the KHS occurs. BR will be maximum (1) if the geographical distributions of KHS are non overlapping. Low values of BR, suggest the parasite has a limited biogeographic range, arguing for users to increase BEV.

PaCo is able to identify the parasite species that cooccur most frequently on ecologically related host species. PaCo uses the same data as PaNic. To build a model, PaCo requires the following steps (see scheme):

The user selects a fish species (B) from the set of those available (A), which includes all species present in the Fishbase Species Ecology Matrices (>27,400).

Niche dimensions of the selected host (C) are compared to those of the members of the projection host set to compute values of ecological similarity (ES)(D).

ES values between each fish species of the projection set and target host are then weighted by a habitat and/or a geography and/or a phylogenetic filter (F).

A C-score is assigned to each parasite species of the internal DB. The C-score is calculated by summing up all the compatibility values of the hosts (excluding the target host) where the parasite occurs (E).

Parasite species are ranked according to their C-scores (G). As a result, parasites that occur in hosts similar to the target host are ranked high, where similarity is defined by the user.

1) Type the scientific name of a host (genus, species and subspecies) in the textbox. If the host name is missing, incomplete, or misspelled PaCo will return an error page.

2) Select one or more parasite groups to be included in the model.

3)Select the host niche dimensions (max length, growth rate (K), life span, age at maturity, and trophic level) you want to include in the model. By default, all dimensions are included. See Fishbase documentation for specific details on host niche dimensions. Host niche dimensions are used by PaCo to calculate euclidean distances between the target host and all other fish species in the internal database.

4) Apply one or more filters to the projection set. You can choose to weight ES values of each host of the projection set according to the relative overlap between its habitat, geography and phylogeny and those of the target host.

5) Set output preferences by choosing the minimum and maximum number of suggested hosts to output and the minimum percentage of known parasites of the target host to include in the output list.

7) Submit your query.

The results page summarizes the main features of the model, comprising:

- the selected host species;

- the selected host niche dimensions and filters;

- the known parasite species/genera for the selected host and their respective host range (i.e. the number of known hosts according to the internal database);

- the top compatible parasites for the target host (according to the output preferences defined in model setup page) and their respective C-score (see next paragraph for further details on C-score computation);

- the average C-score for all the parasites in the internal DB (belonging to the selected taxon/taxa);

- the average C-score for all the parasites in the internal DB already reported for target host;

- the average C-score for all the parasites in the internal DB not yet reported from the target host;

- a Model Evaluation Value (MEV) which is computed as:
1-(ΔC/Cmax),
where Cmax is the maximum C-score and ΔC is the average difference between Cmax and the C-score of each reported parasite. MEV may range between 0 and 1 (the closer MEV is to 1, the better the model has done at predicting parasites already known to occur from the target host);

- a discrepancy index, which is computed as:
D/N,
where D is the number of unreported parasites which should be moved to have all the reported parasites at the top of the list of C-scores, and N is the total number of unreported parasites. Discrepancy may range between 0 (when all the reported parasites have a C-score higher than that of any unreported parasite) and 1 (when all the unreported parasites have a C-score higher than that of any reported parasite).

To gain a better understanding of the algorithm underlying PaCo, you have to think about what kind of parasite should be most likely to be found on a target host. Such a parasite would be a species parasitizing fish ecologically similar to the target host. It would be a generalist (and therefore without physiological obstacles for parasitizing a particular host). In addition, it would be distributed on hosts with similar geography, phylogeny and habitat affinities as the target host. The
PaCo algorithm can take into account all of these properties, by taking three steps.
First PaCo computes the ecological similarity (ES) between the selected host (H) and all the other fish species in the internal database (h_i) on the basis of the selected niche dimensions. For each host h_i, EC is calculated as:

EC=1/(Euc(H,h_i)+1),

where Euc(H,h_i) is the Euclidean distance between H and h computed using the selected niche dimensions.
Second, PaCo weights the ES values using a habitat and/or a geography and/or a phylogenetic filter.
Third, PaCo assigns to each parasite species a C-score, which is obtained by summing up all the compatibility values of the hosts (excluding the target one) where the parasite occurs.
The application of a filter to a fish species of the host set produces a single value ranging from 0 (maximum effect) to 1 (no effect). Filter values are computed as follows:

- Habitat (HAB):

HAB=(M+B+F)/3,

where M, B and F indicate the respective overlaps in habitat preference (marine, brackish and freshwater) between the target host and the considered fish species.

- Geography (GEO):

GEO=GO/GH,

where GO indicates the number of localities where both the target host and the considered fish species occur, while GH indicates the total number of localities where target host occurs.

- Phylogeny (PHY):

PHY=(c+o+f+g+s)/3N,

where s, g, and f indicate the match between, respectively, order, class, family, genus and species of the target host and of the considered fish species. The inclusion of class and order is optional and the recommended level for PHY is that of family (i.e. PHY=f+g+s)/3N).

An exhaustive discussion on how filters and ecological parameters affect performance of PaCo algorithm is provided by Strona and Lafferty (2012b).

PaCo allows users to search for the optimal set of host niche dimensions and filters for a particular host. The algorithm PaCo uses to perform this task works in a two steps process. In the first step, PaCo defines an optimal set of niche dimensions using a simple genetic algorithm. In the second step, PaCo tries all possible combination of filters to find the best one. Because the fit of each parameter combination is estimated using a model evaluation value(MEV, see 3.2 section), which is computed using the C-scores of parasites in the target host, an optimal model search can be performed only for host species with at least 5 known parasites. If the target host does not satisfy this requirements, PaCo will compute a single model.

Parasite List Generator (PaL) allows users to access the FishPEST internal database. PaL allows access to internal data, providing flexibility and advanced functionalities. PaL is innovative in both its query criteria and output construction. Internal database can be filtered according to parasite features (taxon, area of distribution) and host features (phylogeny, habitat, ecology), giving users the ability to test biogeographic, coevolutionary and ecological hypotheses. Consistent with this purpose, PaL offers freedom in output customization. Users can create lists of parasites per host, lists of hosts per parasite and lists of host/parasite records according to the above filters. Additionally, they can create fully customizable presence-absence matrices to be used for further analyses. For each generated list, PaL provides basic statistical information about the output host/parasite assemblages.

Users can create parasite/host lists and matrices by filtering the internal database according to parasite- and host-related criteria, and particularly: 1) parasite taxonomy; 2) parasite biogeography; 3) host habitat; 4) host family; 5) host ecology (maximum length, growth rate, life span, age at first maturity, trophic level). Each category must be checked to be included in the output (i.e., output list/matrix will be empty if a user does not select at least 1 parasite group, 1 area of parasite distribution and 1 host habitat). By default, all families are included. If a user wants to limit his query to 1 or more particular families, he has to manually check them in the provided list. Similarly, the ecological filter is set by default to intervals wide enough to include all hosts from the internal database. If users want to limit the search to hosts with particular ecological features, they have to change the corresponding minumum and maximum value.
By default, the output is given in form of a results page containing the parasite/host list generated according to user selections. Alternatively, users can choose to save the list to an output text file, or to save it in form of a presence-absence matrix to be used for further analyses. Users can decide to whether or not to include row and column names in the output matrix, and to have cells separated by tabulations or by empty spaces. (

PaL can generate lists of known parasites for a particular host, and lists of known hosts for a particular parasite. To generate a host/parasite list or a parasite/host list, a user is simply required to select one of the two options, and to insert genus and, eventually, species of the host or parasite they are interested in. If a user inserts only a genus name, or if the typed specific name does not match any of those included in the internal database, PaL will return a list of hosts/parasites for the selected genus. If the genus does not match any of those present in the internal database, PaL will return an error message.

FishPEST offers researchers a rich set of statistical tools able to enlighten many aspects of host/parasite relationships. We hope that its simplicity and friendly interface makes it a valuable instrument for a non-academic audience, allowing an occasional user to hypothesize what is going on in their home aquarium, or helping a fish farmer to preventively protect its semi-intensive farm from potential parasitic diseases. FishPEST is a non-commercial scientific project, and all the databases have been collected, properly formatted, and validated by its developers. FishPEST makes working with existing data fast and easy, and we hope it can serve Parasitology in the manner that Fishbase serves Ichthyology. The value of FishPEST is limited to the data entered into it. FishPEST users can share their data with fish parasitologist community through FishPEST. Submitting data to FishPEST is very easy. Uploading does not does not require registration. Simply fill out the required fields or submit a properly formatted text document. None of the fields are mandatory, but original data (i.e. data not supported by peer-reviewed literature) will not be taken into consideration unless the uploader provides an e-mail address to be contacted for further discussion. Data providers will be properly credited in the FishPEST contributors page.

Common errors that could occur while creating a model are handled separately by PaNic, PaCo and PaL. In case of incorrect input, users will be redirected to specific error page showing detailed information about what possibly went wrong. Although FishPEST has been intensively tested, we are interested in feedback about bugs or errors in the internal database.