Animal diseases in Switzerland¶
FOAG, Federal Office for Agriculture, collects data on the animal diseases in Switzerland. This data is published as Linked Data.
In this tutorial, we will show how to work with Linked Data. Mainly, we will see how to work with data on animal epidemics.
We will look into how to query, process, and visualize it.
1. Setup
1.1 Python
1.2 SPARQL endpoints
1.3 SPARQL client
2. Data
2.1 Animal species
2.2 Diseases
2.3 Can we link disease to animal type?
2.4 Reports
2.5 Example: goats
3. Analysis
3.1 Cattle: Mucosal Disease
3.2 Bees: Sauerbrut
3.3 Rabies
3.4 Biggest epidemics
3.5 Epidemics over time
Setup¶
Python¶
This notebook requires Python 3.9 or later to run.
SPARQL endpoints¶
For epidemiological data¶
Reports on animal diseases are published as Linked Data. It can be accessed with SPARQL queries.
You can send queries using HTTP requests. The API endpoint is https://lindas.admin.ch/query/.
For geodata¶
Animal diseases are closely linked to places. To understand their location, we will work with Swiss geodata. It is published as Linked Data. It can be accessed using API endpoint under https://ld.geo.admin.ch/query.
import datetime
import itertools
import json
import folium
import mapclassify
import matplotlib as mpl
import matplotlib.cm
import networkx as nx
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from bokeh.io import output_notebook, show
from bokeh.models import (Circle, CustomJS, HoverTool, LabelSet, MultiLine,
NodesAndLinkedEdges, Plot, TapTool)
from bokeh.palettes import Paired
from bokeh.plotting import ColumnDataSource, from_networkx
from dateutil.relativedelta import relativedelta
from folium.plugins import TimeSliderChoropleth
from graphly.api_client import SparqlClient
from plotly.subplots import make_subplots
from enum import Enum
from collections import namedtuple
# Uncomment to install dependencies in Colab environment
#!pip install mapclassify
#!pip install git+https://github.com/zazuko/graphly.git
sparql = SparqlClient("https://int.lindas.admin.ch/query")
geosparql = SparqlClient("https://ld.geo.admin.ch/query")
sparql.add_prefixes({
"schema": "<http://schema.org/>",
"cube": "<https://cube.link/>",
"admin": "<https://schema.ld.admin.ch/>",
"skos": "<http://www.w3.org/2004/02/skos/core#>",
"disease": "<https://environment.ld.admin.ch/foen/animal-pest/>"
})
geosparql.add_prefixes({
"dct": "<http://purl.org/dc/terms/>",
"geonames": "<http://www.geonames.org/ontology#>",
"schema": "<http://schema.org/>",
"geosparql": "<http://www.opengis.net/ont/geosparql#>",
})
SPARQL queries can become very long. To improve the readibility, we will work wih prefixes.
Using the add_prefixes method, we can define persistent prefixes.
Every time you send a query, graphly will now automatically add the prefixes for you.
Data¶
Our goal is to understand outbreaks of animal diseases. First, let's take a look at the animals found in our dataset.
Animal species¶
query = """
SELECT DISTINCT ?specie ?group
WHERE {
?s disease:animal-specie ?specieIRI.
?specieIRI schema:name ?specie.
?specieIRI skos:broader/schema:name ?group.
FILTER (LANG(?specie) = "de")
FILTER (LANG(?group) = "de")
}
ORDER BY ?group
"""
df = sparql.send_query(query)
df[0:15]
Diseases¶
Now, let's take a look at the diseases the animals can suffer from.
query = """
SELECT DISTINCT ?epidemics ?group
WHERE {
?s disease:epidemics ?epidemicsIRI.
?epidemicsIRI schema:name ?epidemics.
?epidemicsIRI skos:broader/schema:name ?group.
FILTER (LANG(?epidemics) = "de")
FILTER (LANG(?group) = "de")
}
ORDER BY ?group
"""
df = sparql.send_query(query)
df.head()
Can we link disease to animal type?¶
Thus far, we have seen all animals, and all diseases. However, not all animals will catch all diseases.
What diseases are swiss animals exposed to?
query = """
SELECT DISTINCT ?epidemics ?specie ?group
WHERE {
<https://environment.ld.admin.ch/foen/animal-pest/observation/> cube:observation ?obs .
?obs disease:epidemics/schema:name ?epidemics;
disease:animal-specie ?specieIRI.
?specieIRI schema:name ?specie.
?specieIRI skos:broader/schema:name ?group.
FILTER (LANG(?epidemics) = "de")
FILTER (LANG(?specie) = "de")
FILTER (LANG(?group) = "de")
}
ORDER BY ?specie
"""
df = sparql.send_query(query)
df = df[df.group != "Wildtier"]
# Graph of epidemics-specie relations
graph = nx.from_pandas_edgelist(df, source='epidemics', target='specie')
groups = {key: "epidemics" for key in df.epidemics} | {key: "specie" for key in df.specie}
nx.set_node_attributes(graph, groups, name="group")
colormaps = {"fill_color":
{"epidemics": Paired[4][0], "specie": Paired[4][2]},
"hover_color":
{"epidemics": Paired[4][1], "specie": Paired[4][3]}
}
for name, colormap in colormaps.items():
colors = {k: colormap[v] for k, v in groups.items()}
nx.set_node_attributes(graph, colors, name=name)
# Interactive graph plot
plot = Plot(plot_width = 1000, plot_height=1000)
plot.title.text = 'Swiss animals and their diseases'
graph_renderer = from_networkx(graph, nx.drawing.layout.bipartite_layout, nodes = df.specie.unique())
# manipulating nodes
nonselect_node = Circle(size = 15, fill_color = "fill_color")
select_node = Circle(size = 15, fill_color = "hover_color")
graph_renderer.node_renderer.glyph = nonselect_node
graph_renderer.node_renderer.nonselection_glyph = nonselect_node
graph_renderer.node_renderer.selection_glyph = select_node
graph_renderer.node_renderer.hover_glyph = select_node
graph_renderer.node_renderer.data_source.data['group'] = [groups[g] for g in graph_renderer.node_renderer.data_source.data["index"]]
# manipulating edges
select_edge = MultiLine(line_color = Paired[7][-1], line_width = 2)
graph_renderer.edge_renderer.glyph = MultiLine(line_color = '#CCCCCC', line_alpha = .5, line_width = 1)
graph_renderer.edge_renderer.selection_glyph = select_edge
graph_renderer.edge_renderer.hover_glyph = select_edge
graph_renderer.selection_policy = NodesAndLinkedEdges()
graph_renderer.inspection_policy = NodesAndLinkedEdges()
plot.renderers.append(graph_renderer)
callback_code = '''
if (cb_data.index.indices.length > 0) {
const index = cb_data.index.indices[0];
if (source.data.group[index] === "specie") {
hover.tooltips = [["Animal", "@index"]];
}
else {
hover.tooltips = [["Disease", "@index"]];
}
}
'''
hover = HoverTool(
renderers=[graph_renderer.node_renderer]
)
hover.callback = CustomJS(
args = dict(source = graph_renderer.node_renderer.data_source, hover = hover),
code = callback_code
)
pos = {k: v for k, v in graph_renderer.layout_provider.graph_layout.items() if graph.nodes[k]["group"] == "specie"}
names = list(pos.keys())
x, y = map(list, zip(*pos.values()))
source = ColumnDataSource({'x': x, 'y': y, 'name': names})
labels = LabelSet(x_offset=15, y_offset=-6, x='x', y='y', text='name', source=source, background_fill_color='white', text_font_size='10px', background_fill_alpha=.7)
plot.renderers.append(labels)
plot.add_tools(hover, TapTool())
output_notebook()
show(plot)
Reports¶
When animals get sick, the farmer will typically consult a vet. After diagnosis, the incident is reported to the Federal Office for Agriculture. This data is publicly available and can be found in our endpoint:
#Run in browser: https://s.zazuko.com/24fSKE
query = """
SELECT ?diagnosis ?commune ?specie ?stock ?sick ?infected ?killed ?deceased ?epidemics
WHERE {
<https://environment.ld.admin.ch/foen/animal-pest/observation/> cube:observation ?obs .
?obs disease:epidemics/schema:name ?epidemics;
disease:diagnosis-date ?diagnosis;
disease:animals-stock ?stock;
disease:animals-sick ?sick;
disease:animals-infected ?infected;
disease:animals-killed ?killed;
disease:animals-deceased ?deceased;
disease:internet-publication ?date;
disease:animal-specie/schema:name ?specie;
schema:containedInPlace/schema:name ?commune .
FILTER (LANG(?epidemics) = "de")
FILTER (LANG(?specie) = "de")
}
ORDER BY DESC(?diagnosis) ?commune
"""
df = sparql.send_query(query)
df.head()
Aggregated reports¶
The reports give us many details. It tells us not only what specie was affected, but also how many animals got sick, infected, or deceased.
These reports, however, have one limitation. They are sent only once per incident. If a goat gets sick, the vet will examine the entire herd. S/he will check how many goats are sick, or infected. S/he may recommend killing parts or the entire herd. The farmer, or the vet, will then report how many aniamls were sick, infected, killed, or deceased.
But that will happen only once. Nobody will examine this herd again, be it the same week or later. Hence we do not know how a given disease evolved. A seemingly insignificant report may turn out to be a serious issue a few days later. We also do not know at which point in time the report was made. Was it a day, two or ten after the first infection?
The data from the reports raises many questions. To address them, we will work with number of reports. Instead of looking at how individual farms were affected, we will simply count how many farms reported health issues.
Our core data becomes:
query = """
SELECT ?diagnosis ?municipality_id ?municipality ?specie ?epidemics
WHERE {
<https://environment.ld.admin.ch/foen/animal-pest/observation/> cube:observation ?obs .
?obs disease:epidemics/schema:name ?epidemics;
disease:diagnosis-date ?diagnosis;
disease:internet-publication ?date;
disease:animal-specie/schema:name ?specie;
schema:containedInPlace ?municipality_iri.
?municipality_iri schema:name ?municipality ;
schema:identifier ?municipality_id.
FILTER (LANG(?epidemics) = "de")
FILTER (LANG(?specie) = "de")
}
ORDER BY DESC(?diagnosis) ?municipality
"""
df = sparql.send_query(query)
def aggregate_reports(reports: pd.DataFrame, weekly: bool) -> pd.DataFrame:
df = reports.copy()
if weekly:
df["diagnosis"] = df.diagnosis.apply(lambda x: datetime.date(x.year, 1, 1) + relativedelta(weeks=+x.isocalendar()[1]))
else:
df['diagnosis'] = df.diagnosis.apply(lambda x: datetime.date(x.year, x.month, 1))
return df[["diagnosis", "specie", "epidemics", "municipality_id"]].groupby(by=["diagnosis", "specie", "epidemics", "municipality_id"]).size().reset_index(name="reports")
# Aggregate results by week/month
df_monthly = aggregate_reports(df, False)
df_monthly = df_monthly[["diagnosis", "specie", "epidemics"]].groupby(by=["diagnosis", "specie", "epidemics"]).size().reset_index(name="reports")
df_monthly = df_monthly.sort_values(by=["diagnosis", "epidemics", "specie"], ascending=False).reset_index(drop=True)
df_monthly.head()
Example: goats¶
We have seen that one animal can suffer from many diseases. Remember the animal-disease graph? A goat can get sick of twenty different diseases!
Goat diseases
Some of them are more serious than others. Which ones were the most often reported in the past?
DimensionData = namedtuple('Dimension', ['slice_of', "category"])
class Dimension(Enum):
Epidemic = DimensionData("epidemics", "specie")
Specie = DimensionData("specie", "epidemics")
def plot_multiple_categories(df: pd.DataFrame, dim: Dimension, value: str, min_reports: int=15, min_reports_at_time:int = 5) -> None:
category = dim.value.category
df_slice = df[df[dim.value.slice_of] == value]
counts = df_slice[[category, "reports"]].groupby(category).agg(["sum", "max"]).reset_index()
categories_filtered = list(counts[(counts["reports"]["sum"] >= min_reports) & (counts["reports"]["max"] >= min_reports_at_time)][category])
df_slice = df_slice[df_slice[category].isin(categories_filtered)]
all_dates = [x.date() for x in pd.date_range(df_slice.diagnosis.min(), df_slice.diagnosis.max(), freq='MS')]
combinations = list(itertools.product(*[all_dates, df_slice[category].unique()]))
values_empty = pd.DataFrame(combinations, columns =['diagnosis', category])
plot_df = pd.merge(values_empty, df_slice, how="left", on=["diagnosis", category]).fillna(0)
fig = go.Figure()
for cat in plot_df[category].unique():
plot_df_slice = plot_df[plot_df[category] == cat]
labels = ["<b>{}</b><br>Date: {}<br>Reports: {}".format(cat, row.diagnosis, int(row.reports)) for row in plot_df_slice[["diagnosis", "reports"]].itertuples()]
fig.add_trace(go.Scatter(x=plot_df_slice.diagnosis, y=plot_df_slice.reports, name=cat,
text=labels,
hoverinfo='text',
line_shape='spline'))
fig.update_xaxes(rangeslider_visible=True, range=[min(plot_df.diagnosis), max(plot_df.diagnosis)])
fig.update_yaxes(title="reports", range = [0, int(max(plot_df.reports)*1.1)+1])
fig.update_layout(
title={"text": "Epidemics: {}".format(value), "x": 0.5}
)
fig.show()
plot_multiple_categories(df_monthly, Dimension.Specie, "Ziege", 15, 5)