As the Covid-19 pandemic has shown, we live in a highly connected world that enables not only the efficient spread of a virus, but also information and influence. What can we learn from analyzing these connections? This is a core question in network science, an area of ââresearch that models interactions between physical, biological, social, and information systems to solve problems.
The 2021 Graph Exploitation Symposium (GraphEx), hosted by the MIT Lincoln Laboratory, brought together leading network scientists to share the latest advances and applications in the field.
“We are exploring and identifying how leveraging graph data can provide key technologies to solve the most pressing problems our nation faces today,” said Edward Kao, symposium organizer and technical staff member at Lincoln Laboratory Laboratory AI software architectures and algorithms group.
The virtual event’s themes revolved around some of the year’s top themes, such as analyzing disinformation on social media, modeling the spread of the pandemic, and using graph-based machine learning models to accelerate drug design.
“The Influence Operations and Covid-19 Special Sessions at GraphEx reflect the importance of network and graph-based analysis in understanding the phenomenology of these intricate and influential aspects of modern life and may also suggest ways as we learn more and more about manipulation of graphs “says William Streileinwho co-moderated the event Rajmonda Caceres, both from Lincoln Laboratory.
Several presentations at the symposium focused on the role of network science in analyzing influence operations (IO) or organized attempts by state and / or non-state actors to spread disinformation narratives.
Lincoln Laboratory researchers developed tools Classify and quantify the impact of social media accounts that are likely to be IO accounts, such as: B. those who deliberately spread false Covid-19 treatments to vulnerable population groups.
âA cluster of IO accounts acts as an echo chamber to amplify the narrative. The population at risk then participates in these narratives, âsays Erika Mackin, a researcher who uses the tool called RIO or. developed Clarification of influence operations.
To classify IO accounts, Mackin and her team trained an algorithm to identify likely IO accounts on Twitter networks based on a specific hashtag or narrative. One example they examined was #MacronLeaks, a disinformation campaign targeted against Emmanuel Macron during the 2017 French presidential election. The algorithm is trained to mark accounts within this network as IO based on several factors, such as the number of interactions with foreign message accounts, the number of tweeted links, or the number of languages ââused. Your model then uses a statistical approach to assess an account’s impact on the spread of the narrative within that network.
The team found that their classifier outperforms existing detectors of IO accounts as it can identify both bot accounts and human operated ones. They also found that IO accounts that drove the disinformation narrative of the 2017 French elections largely overlapped with accounts that are now spreading influential disinformation about the Covid-19 pandemic. “This suggests that these reports will continue to move to disinformation narratives,” says Mackin.
Throughout the Covid-19 pandemic, leaders have looked for epidemiological models that predict the spread of disease in order to make informed decisions. Alessandro Vespignani, Director of the Network Science Institute at Northeastern University, led the Covid-19 modeling effort in the United States and gave a keynote address on that work at the symposium.
In addition to taking into account the biological facts of the disease, such as the incubation period, Vespignani’s model is particularly strong in the inclusion of community behavior. In order to conduct realistic simulations of the spread of the disease, he is developing “synthetic populations” which are created from publicly available, highly detailed data sets on US households. âWe create a population that is not real, but statistically real, and create a map of the interactions between these individuals,â he says. This information flows back into the model to predict the spread of the disease.
Vespignani is now considering how genomic analysis of the virus can be integrated into this type of population modeling to understand how variants spread. “It’s still very interesting work,” he says, adding that this approach was useful in modeling the spread of the delta variant of SARS-CoV-2.
As researchers model the spread of the virus, Lincoln Laboratory’s Lucas Laird ponders how network science can be used to develop effective control strategies. He and his team develop a model to customize strategies for different geographic regions. Efforts have been spurred by the differences in Covid-19 that are prevalent in U.S. communities and the gap in intervention modeling to address those differences.
As examples, they applied their planning algorithm to three counties in Florida, Massachusetts, and California. Taking into account the characteristics of a particular geographic center, such as the number of people at risk and the number of infections there, your planner will implement different strategies in those communities throughout the outbreak period.
“Our approach eliminates disease in 100 days, but is also able to do it with much more targeted interventions than any global intervention. In other words, you don’t have to shut down an entire country.” Laird adds that her planner provides a “sandbox environment” to explore future intervention strategies.
Machine learning with graphics
Graph-based machine learning is receiving increasing attention for its potential to âlearnâ the complex relationships between graphical data and thus gain new insights or predictions about these relationships. This interest has led to a new class of algorithms called neural graph networks. Today, graphic neural networks are used in areas such as drug research and material design with promising results.
âWe can now apply deep learning much more broadly, not just to medical images and biological sequences. This creates new opportunities in data-rich biology and medicine, âsays Marinka Zitnik, an assistant professor at Harvard University who presented her research at GraphEx.
Zitnik’s research focuses on the rich networks of interactions between proteins, drugs, diseases, and patients, spanning billions of interactions. One application of this research is the discovery of drugs to treat diseases with little or no approved drug treatments, such as Covid-19. In April, Zitnik’s team published an article about their research Using the Neural Graph Network to rank 6,340 drugs for expected effectiveness against SARS-CoV-2 and identify four that could be repurposed to treat Covid-19.
Similarly, at Lincoln Laboratory, researchers are using graphene neural networks to develop advanced materials, such as those that can withstand extreme radiation or trap carbon dioxide. Like the drug development process, the trial and error approach to material design is time consuming and expensive. The laboratory team develops neural graph networks that can learn the relationships between the crystalline structure of a material and its properties. This network can then be used to predict a wide variety of properties of any new crystal structure, greatly accelerating the screening of materials with desired properties for specific applications.
“Learning graph representations has become a rich and thriving area of ââresearch to incorporate inductive biases and structured priors during machine learning, with broad applications such as drug design, accelerated scientific discovery, and personalized recommendation systems,” says Caceres.
A living community
Lincoln Laboratory has hosted the GraphExp Symposium annually since 2010, with the exception of the cancellation last year due to Covid-19. âOne important finding is that, despite the shift from last year and the need to be virtual, the GraphEx community is more lively and active than ever before,â says Streilein. “Network-based analysis continues to expand its reach and is increasingly applied to increasingly important areas of science, society and defense.”
Members of the technical committee and co-chairs of the GraphEx Symposium included researchers from Harvard University, Arizona State University, Stanford University, Smith College, Duke University, the Department of Defense, and the US Department of Defense, in addition to Lincoln Laboratory staff Sandia National Laboratories.