Caroline Casey - Week 6 - Complex Systems Group: Connector Hubs, Participation Coefficients, and More Writing - University of Pennsylvania

This week was exciting as I navigated my project completely by myself! Dr. Bassett was on vacation this week, so it was up to me to figure out what analyses to complete and how to complete them. I feel very comfortable with Matlab as well as network science and managed to complete a lot this week.

On Monday morning, I added what I worked on the previous week to my PowerPoint. I also began looking at the communities I identified in order to determine the next step in the process of analyzing the networks. Although Dr. Bassett was on vacation this week, she came back for our lab meeting. In the meeting, one of the post-docs in the lab presented his work on temporal (time dependent) brain networks and how they change when completing memory tests. It was a very interesting lecture and I felt like I could understand most of the concepts. After the lecture, each member of the lab gave Dr. Bassett a five minute update on his/her completed work from the past week. I enjoyed hearing what everyone in the lab is working on; the projects are all very diverse. Dr. Bassett suggested that for this week I look more into the communities I identified and anything else I find important. I ended Monday drafting ideas for how I would go about analyzing the communities I found and what interesting data and results I could pull from the networks.

On Tuesday, I performed the consensus partitions multiple times again for each of the four scenarios because not all of the 100 optimizations were yielding the same community partitions for all 112 nodes. However, the consensus partitions came back the same as before, there was still disagreement on the community partitions for some of the nodes. So, I took the community each node appeared in most often and let that be the community assignment for each node. Next, I created four vectors, 1x112, which contained the community assignment of each node for each scenario. I then plotted the resulting nodes, where the color of the node indicated which community it was assigned to. I created and saved the networks and then plotted the nodes onto a brain surface for each scenario. I analyzed the resulting networks for trends among the four scenarios in order to identify whether or not certain nodes tended to appear in a community together. Unfortunately, there were no obvious trends. I added each of the images and explanations of what I did to the PowerPoint. I then added the community analyses I completed to the methods and results sections of the paper I am writing.

The brain surface plots of the first scenario (edges from the correlation of interference session 2 with scan 1). The color of the node indicates which community it is assigned to. 

Wednesday was devoted to reading literature on network analysis methods and graph theory. I was looking for other methods of analyzing the networks. By Wednesday afternoon, I had two analyses techniques that I wanted to complete, finding the betweenness centrality of each node and the participation coefficient of each node. Betweenness centrality is a fraction explaining the number of all shortest paths in the network that pass through a certain node. A high betweenness centrality indicates that the specific node connects various parts of the network together (it plays a very central role). Nodes on the edges of networks tend to have a very low betweenness centrality. The participation coefficient helps identify which nodes connect to other communities and which nodes only connect within their own communities. A participation coefficient close to 1 indicates that the connections from that node are evenly distributed among all the communities. A participation coefficient of 0 indicates that the connections from that node are all within its own community. I spent the rest of Wednesday finding the algorithms and figuring out how to employ the two analyses techniques.

On Thursday, I began by finding the betweenness centrality of each node for all of the networks. After obtaining the values, I then found the connector hubs of each of the networks. Connector hubs are nodes whose betweenness centrality is greater than the mean plus the standard deviation. These nodes connect various parts of the network together and play a very central and global role. I identified these nodes and then added them to an Excel sheet. I compared the connector hubs of the two scenarios in the predictive group (edges from the correlation of interference session 2 with scan 1 and the edges from the correlation of interference session 3 with scan 2) to see if there were any regions that were hubs in both scenarios. I then did the same for both scenarios in the retrospective group (edges from the correlation of interference session 2 with scan 2 and edges from the correlation of interference session 3 with scan 3). I found some conserved regions and added my work to the PowerPoint. Thursday afternoon, I began working on finding the participation coefficient of each of the nodes.

Friday, I finished finding the participation coefficients of the nodes. I then found the node with the highest participation coefficient in each of the four scenarios and then found the nodes with the lowest participation coefficients (a participation coefficient of 0) for each of the four scenarios. I compared the two scenarios of the predictive group and compared the two scenarios of the retrospective group to find conserved regions. I believe that there is more I can do with the participation coefficients, but I will talk to Dr. Bassett on Monday about her suggestions. I then added my work to the PowerPoint and continued working on my paper. As I was going through my paper, I realized that last week when I was computing the z-score (to find the hubs in the networks) and then comparing the two scenarios in the predictive group and the two scenarios in the retrospective group, I did not necessarily find all hubs in common. This is because in my Excel sheet of hubs for each scenario, I separated the left side of the brain from the right. For example, if the left superior frontal gyrus was a hub in one scenario and then the right superior frontal gyrus was a hub in the other scenario, it was not identified as a conserved hub region for that group. So, I decided to go back and identify all the conserved hubs in both groups regardless of whether the hub was on the right side of the region for one scenario and on the left side of the region for the other scenario. This resulted in many more conserved hubs. I then went on to research the regions of the connector hubs (from the betweenness centrality) and the hubs from the z-score that were conserved between the two scenarios in the two groups (predictive and retrospective). Something interesting I noticed was that a majority of those regions had visual or environmental functions such as face recognition, object recognition, color recognition, spatial orientation, and self-awareness. I think that this finding might indicate that being in a different/unknown environment might be a cause of the interference effect, however, this is just a hunch. I will talk to Dr. Bassett on Monday about her take on the meaning of these results.

On Friday, August 1, I will have a Skype conference with two other researchers where I will share my PowerPoint and discuss my work and results. I enjoy how my project is at the cross-roads of so many fields such as biology (specifically neurology), graph theory, and computer science. Not to mention, I am also gaining skills in making presentations and writing research papers! So far, I am loving the experience in my lab and really enjoy the work that I am doing!

My desk at the lab.