Slime Mold Graph Repository

<h1>The KIST Europe Data Set</h1>

<h2>Introduction</h2>

<p>
This dataset focuses on networks formed by <strong>Physarum Polycephalum</strong>. Detailed methods are available in the <a href="https://doi.org/10.1088/1361-6463/aa7326">companion paper</a>.
</p>

<h2>Description</h2>

<p>The KIST Europe data set contains raw and processed data of 81 identical experiments, carefully executed under
constant laboratory conditions. The data was produced using the following procedure:</p>

<ol>
<li>A rectangular plastic dish is prepared with a thin sheet of agar.</li>
<li>A small amount of dried <em>P. Polycephalum</em> (HU195xHU200) <b>sclerotia crumbs</b> is lined up along the
short edge of the dish. The dish is put into a large light-proof box.
</li>
<li>After approximately <b>14 hours</b> the plasmodium has resuscitated and starts exploring the available space
towards the far side of the dish. Typically, the apical zone needs to cover a distance of several
centimeters before network formation can be observed properly.
</li>
<li>For the next <b>30 hours</b> we take a <b>top-view image</b> of the growing plasmodium and its changing
network every <b>120 seconds</b> from a fixed position. We stop capturing when the apical zone is about to reach the far side of the dish, which is
outside of the observed area.
</li>
<li>After obtaining <b>sequences of images</b> showing the characteristic networks of <em>P. Polycephalum</em>,
we use a software called <a href="http://nefi.mpi-inf.mpg.de">NEFI</a> to compute corresponding <b>sequences
of graph representations</b> of the depicted structures within a predefined region of interest.
In addition to the topology the graphs store precise information of the <b>length
and width</b> of the edges as well as the <b>coordinates</b> of the nodes in the plane.
</li>
<li>Given the resulting sequence of graphs we apply <b>filters</b> removing artifacts and other unwanted
features of the graphs. Then we proceed to compute a <b>novel node tracking</b>, encoding the time
development of every node, taking into account the changing topology of the evolving graphs.
</li>
</ol>

<p>Repeating this experiment we obtain <b>81 sequence of images</b>, which we consider our <b>raw data</b>. We
stress at this point that given the inherently uncontrollable growth process of <em>P. Polycephalum</em>, the
obtained sequences differ in length and nature. That is to say, in some experiments the organism behaved
unfavorably, simply stopping its growth, changing direction or even escaping the container. While such sequences
are part of the raw dataset, we excluded them partially or completely from the subsequent graph extraction
efforts. The removal of such data reduces the number of series depicting proper network formation to <b>54</b>.
</p>

<p>After obtaining the raw data, we transform the images into <b>equivalent mathematical graphs</b>, thus opening up
a wealth of possibilities for data analysis. To this end we deploy a convenient automatic software tool called
<a href="http://nefi.mpi-inf.mpg.de">NEFI</a>, which analyzes a digital image, separates the depicted slime mold
network from the background and returns a graph representation of said structure. Using this tool effectively
requires some moderate amount of image preprocessing. In particular, for each sequence of images it is necessary
to decide on a suitable <b>subsequence</b> to be processed. Here we typically exclude parts of the sequence
where the apical zone is still visible. For each such subsequence a suitable <b>region of interest</b> is
defined manually. The graph stores the
position of the nodes in the plane as well as edge attributes such as edge length and widths for each edge. In
addition to the output of NEFI including the unfiltered graphs, the dataset contains NEFI's input, i.e. the
selected subsequences of images cropped according to their defined regions of interest.</p>

<p>Note that some parts of the image series showing proper network formation did not yield optimal representations
of the depicted networks. This is a result of images exhibiting strong color gradients rendering them too
challenging for automatic network extraction. While such cases can still be handled by tuning the parameters of
image processing manually on an image per image basis, we decided to discard affected series from subsequent
processing efforts. As a result the number of <b>usable graph sequences of highest quality</b> reduced to
<b>36</b> to which we apply a set of filters removing artifacts, isolated small components and dead-end paths.
Thus we obtain a total of <b>3134 distinct filtered graphs</b> faithfully reflecting the topology and edge
attributes that <em>P. Polycephalum</em> displayed during the wet-lab experiments. At this point available graph
analysis packages or custom written analysis code can be deployed to investigate the data in various ways. The
dataset includes the filtered graphs as well as all corresponding graph drawings. The latter enable a <b>quick
visual inspection</b> of the results of the graph extraction.</p>

<p>Given the obtained time-ordered sequences of graphs the development of the entire graph can be investigated.
However, one may also study what happens to single nodes as <em>P. Polycephalum</em> evolves. Given a graph in a
time ordered sequence of graphs, let us pick any node <em>u</em>. Can we find pick a set of nodes from graphs in
the sequence that are equivalent to <em>u</em>, that is, all nodes in the set are earlier or later versions of
<em>u</em> with respect to time? To answer this question we compute a so-called <b>node tracking</b> which
establishes the time development of all nodes in the graph. Crucially this tracking takes into account
topological changes in the evolving graphs. The result of the tracking is stored as node properties of the
graphs. Naturally, the program computing the tracking is include in the dataset. To the best of our knowledge,
this type of data is made available for the first time through the KIST data set.</p>

<p>Finally, in addition to the actual data, i.e. images and graphs, the KIST Europe data set contains scripts and
larger programs used to process and evaluate the data. Suitable configuration files specify the used regions of
interest and the parameters used with <a href="http://nefi.mpi-inf.mpg.de">NEFI</a>. Thus it becomes possible to
repeat the entire data production process from the raw images to the obtained filtered graphs including the
tracking of nodes.</p>