Fig 1.
A) Timeline of the experiment. B) Graphical representation of the video recording setup. 30 flies were recorded for 10 minutes in a 120 mm diameter circular arena with a bottom made of white Plexiglas illuminated with a white LED strip. C) Cross section of the arena. The outer cover of the arena is made of transparent Plexiglas placed at a height of 3 mm, to allow movement of flies in two dimensions only. D) Distance interaction threshold and possible types of interactions between two flies extracted from the video recording shown in the circle. The size of one fly is considered as one body length (1b) which is approximately 2 mm. An interaction threshold of two body lengths (2b) or 4 mm was used where one fly (the interactor fly shown in color) approaches another fly (the interacted fly shown in gray) in various orientations. Possible interactions are: head-to-head (violet arrow), head-to-tail (yellow arrow) or head-to-body (blue arrow). Data were extracted using Flytracker, which is written in MATLAB. Measured interactions are not oriented, but there is an additional threshold of the length of the interaction of a minimum of 0.5 seconds. E) Data extracted using Flytracker was used to analyze and visualize social interactions between flies at a local, middle, and global level. A package, written in the Python programming language, was used to analyze network data. For visualization and editing, we used the open-source software Gephi. All data and the Python package is publicly available at https://github.com/milanXpetrovic/my_module.
Fig 2.
Visual representation of differences between local network-based measures in terms of the node centrality. Red colored nodes are nodes with the highest value of the centrality measure, while blue colored are nodes with lower values of the centrality measure. This property varies and depends on the chosen centrality network measure.A) Degree centrality or number of individual interactions (links) with other flies (nodes) in the network. B) Closeness centrality measure defines how close a fly (node) is to all other flies (nodes) in the network. C) Betweenness centrality detects the influence of a fly (node) on the flow of information, depicted as a bridge from one part of a network to another. D) Eigenvector centrality measures the influence a fly (node) in a network based to its neighbours. E) Information centrality or average information of all paths originating from a fly (node). F) Clustering coefficient defines how well are neighbouring flies interconnected.
Table 1.
Global level network-based measures.
Table represent values calculated at the global level of network for control (CTRL) and cocaine (COC) populations. Usage of specific edge weight (count, duration) in the calculation of measure is indicated in brackets next to the name of the measure.
Fig 3.
COC and CTRL heat maps of location of interactions.
Visualization shows a graphical representation of the retention duration at a particular location in the arena during experiment. The visualization was made for populations in total where A) shows COC population and B) CTRL population.
Fig 4.
Differences between CTRL and COC SINs in degree centrality, closeness centrality, and clustering coefficient.
A), D), G) Box plot of median values for nine CTRL SINs (n = 270 flies) grouped on the regular food and eleven COC SINs (n = 330) grouped and orally administrated to 0.50 mg/mL of cocaine for 24 hours before tracking in degree centrality A), closeness centrality D) and clustering coefficient G). Data are extracted from 10 minute videos using FlyTracker and analyzed using NetworkX. Statistical analysis was performed using independent-samples t-tests with Welch-correction since group sizes are different. p- values less than 0.05 are taken as significant and presented in S1 Table. B), E), H) Graphical illustration of local-level SINs from the isolated CTRL group using an open-source software Gephi for visualization and editing of the visual appearance of a given network. Nodes are represented with size and color proportional to the values of their centrality measures. Nodes with higher values of centrality measure degree centrality B), closeness centrality E) and clustering coefficient H) are shown as bigger and darker. C), F), I) Graphical illustration of local-level SINs from the isolated COC group using an open-source software Gephi for visualization and editing of the visual appearance of a given network. Nodes are depicted with the size and color proportional to the values of their centrality measures. Nodes with higher values of centrality measure degree centrality C), closeness centrality F) and clustering coefficient I) are shown as bigger and darker.
Fig 5.
Information centrality indicate smaller number of hubs in CTRL networks.
A) Box plot of median values for nine CTRL networks (n = 270 flies) grouped, which were given regular food, and eleven COC networks (n = 330 flies) grouped, which were orally administrated 0.50 mg/mL of cocaine for 24 hours before tracking. Data are extracted from 10 minute videos using FlyTracker and analyzed using NetworkX. Statistical analysis was performed using independent-samples t-tests with Welch-correction since group sizes are different. p-value less than 0.05 is taken as significant presented in S1 Table. B) Box plot of nine CTRL networks (n = 270 flies) grouped on the regular food and eleven COC networks (n = 330 flies) grouped and orally administrated to 0.5 mg/mL of cocaine for 24 hours before tracking. Data are extracted from 10 minute videos using FlyTracker and analyzed using NetworkX. C) Histogram for average information centrality from nine CTRL networks (n = 270 flies) grouped on the regular food. D) Histogram for average information centrality from eleven COC networks (n = 330 flies) grouped and orally administrated to 0.5 mg/mL of cocaine for 24 hours before tracking.
Fig 6.
Node strength distribution using weights uncover more of longer interactions in the COC SINs.
Strength distribution, using weight A) count and B) duration. Box plot of average values for nine CTRL SINs (n = 270 flies) grouped, which were given regular food, and eleven COC SINs (n = 330 flies) grouped, which were orally administrated 0.5 mg/mL of cocaine for 24 hours before tracking. Data are extracted from 10 minute videos using FlyTracker and analyzed using NetworkX. Statistical analysis was performed using two-sample t-tests for a difference in mean for paired samples with a p-value less than 0.05 taken as significant, presented in an S1 Table.
Fig 7.
Image represents correlation of local level measures in created networks. A) is for CTRL networks and B) is for COC networks. Full names of measures that are abbreviated in the correlation matrix figure are: dc—Degree centrality, ce—Eigenvector centrality, cc—Closeness centrality, s—Strength distribution, bc—Betweenness centrality, C—Clustering coefficient and w—weight.
Table 2.
Mid level network-based measures.
Table represent values calculated on middle (community) level of network for control (CTRL) and coaine (COC) population, where values are calculated using duration as node weight.
Fig 8.
Higher number and duration of interaction in denser COC communities.
A) Hypothetical network representation with high modularity B) Hypothetical network representation with low modularity C) Unweighted communities in CTRL D) and COC population E) weighted networks for number of interactions in CTRL F) and COC population, G) weighted networks for duration of interactions for selected CTRL H) and COC group. Each community is represented in different colour and links of different thickness represent the link weight. Ilustrations are created using the open-source software Gephi.