Category: Summer Team Impact Project

Disease spread simulations are critical in evaluating and aiding policy-making during pandemics. However, traditionally, disease spread simulations have poor representations of human mobility. In agent-based models, human mobility is often assumed to be random due to a lack of data. In other types of models, mobility is not included, such as the case in traditional SEIR models, which use differential equations to model disease spread. Furthermore, disease spread simulations constructed by non-experts may have poorly justified disease parameters. In this study, we compare the effect of different representations of mobility in an agent-based model of COVID-19 spread for Fairfax County, Virginia. We use two representations: 1) random mobility and 2) a mobility model that is calibrated using data of foot traffic from SafeGraph. We also compare two different types of agent interaction, the dynamic that drives disease transmission. In one version of the model, an interaction parameter is used to determine the probability that two agents interact at a Place of Interest (POI). Another version of the model uses POI density (square footage per agent) to determine interaction probability. To validate the model, we seed the simulations with empirical case counts of Covid-19 in Fairfax County from April-1-2020 and run the simulation for a year. Our best-fitting simulation, which was one with random mobility, had a Residual Mean Square Error of ~36700 cases compared to simulation versions with empirically-based mobility that had RMSE over 100,000 cases. Our disease parameters, however, replicate average household COVID-19 transmission numbers. Future agent-based models should incorporate mask use and model human risk assessment on compliance with public health guidelines.

00:00 hello everyone my name is fahad aloraini 00:03 and this is our summer team impact 00:05 projects of 00:06 2021 examining different disease 00:09 transmission approaches and data-driven 00:11 agent-based models to reduction 00:17 disease modeling is critical for policy 00:20 making during pandemics 00:21 it helps policy makers make decisions 00:24 predict the outcome of of different 00:27 policies 00:28 and predict a number of cases 00:35 traditionally disease models do not 00:37 include mobility 00:38 such as the case in seir models are 00:41 assumed to be random due to lack of data 00:44 in the current study we create an 00:46 agent-based model with two different 00:48 mobility sub-models 00:49 random or data-driven mobility based on 00:52 safegraph data 00:53 we also create two agents interaction 00:55 models 00:56 one that is interaction parameter based 00:59 and the other 00:59 is density based or determines the 01:02 interaction of 28 based on that 01:05 square footage divided by the number of 01:07 people at that 01:09 location in the simulation at for that 01:12 specific uh time step 01:17 here are the model parameters uh real 01:19 quick interaction probabilities either 01:21 from one percent to fifty percent and 01:24 it’s random and one version of them all 01:26 and another which is a density base the 01:28 version two it’s based on the square 01:30 footage per agent and it uses an 01:32 equation that describes 01:34 fixing chance as a function of test 01:36 distance 01:38 incubation period uh is based on a 01:41 ra it’s random based on a gamma 01:43 distribution 01:45 uh and here are the different 01:49 disease stages 01:53 the transmission probability is based on 01:57 a binomial distribution where n is 02:01 the number of possible interactions at a 02:04 place of interest 02:08 and the dwell times the time it takes or 02:11 a person or an agent stays at a 02:15 location is based on safegraph data 02:18 number of agents is based on the data 02:20 data from uh 02:24 the census and its census data of 02:28 fairfax county in 2018 location choice 02:32 is random or data driven and the random 02:37 as i said it’s people 02:40 or agents choose the locations randomly 02:44 and the data driven model 02:48 they choose the locations based on 02:52 a calibrated model 02:56 based on safe graph data simulation time 02:58 we ran the simulation for 365 03:02 days uh simulation time 03:05 and uh and it takes two and a half days 03:09 to run this 03:10 and the we started 03:13 at the first of april 2021 03:18 2020 verification validation 03:21 code checks and reviews were done 03:23 periodically to verify the model we use 03:25 a residual muse current error to compare 03:27 against covet 19 data 03:28 from the new york times data set 03:32 2021 preliminary results here are our 03:36 results 03:37 and we see that uh 03:41 in many of the simulations a lot of 03:43 people 03:45 become effect infected real quick and 03:49 the simulation stops even before 03:52 the 100th day is reached in other 03:55 simulations uh 03:57 we see that they continue for the whole 03:59 365 days 04:01 uh the steps in those graphs 04:04 are because we average some simulations 04:07 this with same parameters 04:10 because they have a stochastic nature 04:12 they 04:14 are end or terminate 04:18 early because everyone gets infected or 04:20 the infection 04:21 dies out real quickly so you get this 04:25 step function here 04:28 the the line that is well it’s 04:32 it’s easier to see here the green line 04:34 the 04:36 uh at the bottom is the 04:39 empirical data and the other 04:44 lines are the 04:48 simulations and we see that the closest 04:51 thing for the empirical 04:55 is the simulations that are random 04:59 and end early 05:03 here are the aram rmses 05:06 and we see that 36 000 05:10 is our best uh 05:13 simulation in terms of the average 05:16 number of cases it’s all from the 05:18 empirical data 05:20 uh and it’s an interaction based at 05:23 one percent for the 05:26 density based version we see that it is 05:29 37 05:30 000 and both are random mobility 05:34 and we will discuss why in a second 05:37 all other simulations with high 05:39 interaction or lda based mobilities 05:41 had uh rmses of 05:44 100 000 and more they were all from the 05:48 empirical data 05:49 on average 100 000 per day 05:52 uh so the possible reason why 05:56 the empirical based mobility had worst 05:58 fit in comparison with the empirical 06:00 data 06:00 is because it replicates hot spots 06:04 uh 06:08 the the empirical data replicates 06:10 hotspots 06:11 while the random mobility does not 06:13 future work might implement mask use 06:15 this work is still in progress uh so 06:19 this is these results are preliminary 06:21 and we are running 06:22 currently newer versions of the model 06:26 here are my references and here is an 06:29 acknowledgment to 06:30 the nsf the summer team impact 06:35 uh grant of george ma of the jordan 06:37 mason university of office of the 06:39 provost and executive 06:40 vice president this work is supported by 06:43 the george mason university aspiring 06:45 scientist summer internship program 06:48 thank you very much

Analysis of COVID-19 Genome Data in the US Radhika Laddha, Nicole Liang, Minh Tri Le, Taylor Anderson, Amira Roess, Hamdi Kavak, Andreas Zufle Phylogenetic analysis of the COVID-19 virus is vital to identify the various strains, where and when the first case of a strain originated, and the spread dynamics of each strain. We extracted COVID-19 phylogenetic data from the GISAID website for two states in the US, Louisiana and Oregon, and parsed the data into a format that was usable. The data includes the geographic flows of many COVID-19 strains and lineages with origin and destination locations, strain names, recorded dates, and the variants from December 2019 to June 2021. Using various network analysis and visualization tools, we create spatial visualizations of the phylogenetic trees from the data. Future work will enrich these visualizations to examine the connectivity, disease, and sociodemographic characteristics of the regions where new strains emerge. By analyzing thousands of cases of various strains provided by GISAID that include the location and timestamp of each node, we hypothesize that similarities in socioeconomic and geographic factors can be drawn between the mutation originating nodes. Our aim is that our visualizations will aid in communicating the flows of the different strains of the COVID-19 pandemic geographically, identify the relationship between strains and the various characteristics of the regions in which they emerge. Thus, we hope that as a result of our research, we can identify the regions that are at risk of the emergence of new strains. From our results, we can deepen our understanding of COVID-19 in hopes of being better prepared for the next pandemic.

Hey everybody this is Radhika, Nicole, and Tri, and we’re gonna be talking about analyzing the mobility of COVID-19 genome data in the US. COVID-19 has affected the everyday life of billions of people across the world and with new strains emerging and originating in various countries, this virus and its spread has become more uncontrollable and deadly, leading to change in the geographic flows globally. Understanding these changes, identifying patterns can allow for improved health guidelines by informing and communicating with the public, being better prepared for the next pandemic, and knowing how to take more effective action by learning from mistakes. Previous studies have only looked at a particular strain, lineage, or variant of a virus locally or just studied cases in one country. So, for the methods of this research project, we acquired the data set from GISAID and Nextstrain databases for the mobility of COVID-19 virus from 2019 to 2021. We extracted the data from a json format file to a more usable format such as graph and dataframe and analyze it in python. And using various network analysis and visualization tools, we can create spacial visualizations that show the mobility of the virus strain with originating and destination locations. Then we can better understand the flow of the disease throughout about a 2 years period. And then, we used some packages in python and we created some visualizations for the map and for the mobility of the virus. So this map is created using data from the GISAID database, but only from January 202 to April of 2020. And these data points are also only either to the US or from the US. Ok so, a little bit more details about the mobility data. So we have acquired the more detailed data from Louisiana state and Oregon state in the US. This focuses more on two states and small counties within the state we can better understand the various specific regions. These are, as you can see on the map, Louisiana data, and it represents nodes and edges on the graph data. And this is the Oregon data. So by taking these cases from our GISAID data and looking at the different factors, which include location, timestamp, other maybe personal factors about the human such as their sex and age, we can look at the similarities in socioeconomic and geographic data and draw conclusions between the nodes in which new strains are originating and spreading from. By using human mobility data, which we were able to access from a paper by Song Gao, we can also examine the connectivity and disease characteristics and identify the regions are possibly at risk for developing new strains. So, from this, the phylogenetic analysis of the COVID-19 virus, it’s vital to identify the various strains, lineages, and variants, where and when the first case of the strain, lineage, or variant originated, and the spread dynamics of each strain. Our visualizations aid in communicating the flows of the different strains, lineages, and variants of the COVID-19 pandemic geographically. We hope that as a result of our research, better controls, policies, and protocols can be implemented by the World Health Organization and Centers for Disease Control. Preparedness is always key.

There have been many studies on location prediction that have tried to predict the number of visitors at a certain location; however, these studies suffer from some limitations due to the lack of data. This study attempts to create a location prediction system based on data for Fairfax County from SafeGraph. SafeGraph data provides us with a sample of approximately 10 percent of the number of visitors from each census block group to each place of interest from Jan 2018 to Jun 2021. Tensor factorization was used to cancel out the noise that is present in the SafeGraph data. We developed three baseline prediction models to compare against more sophisticated approaches: weekly rolling average [model 1], using previous 4 weeks [model 2], and using previous 4 weeks weighted [model 3]. Other approaches used were long short-term memory (LSTM), regression, croston, autoregressive integrative moving average (ARIMA), and exponential smoothing. The baseline approaches, specifically model 2 and model 3, have consistently performed better than the more sophisticated approaches and tensor factorization has always influenced our models positively. Given that we have access to this abundance of data, using the power of factorization and simple baseline algorithms, businesses and entities can predict human mobility and identify a potential group for their mutual betterment through marketing and advertisements.

our presentation is on spatial temporal prediction of human mobility our research objective is to predict human mobility in fairfax county from a census block group to a place of interest using data from safe graph so we’re going to define points of interest and census block groups a point of interest is basically just any location that shows up on a gps like a hotel or a gas station and a census block group is just a geographical unit used by the census bureau and then to simplify that a little bit more basically we’re looking at someone from cbg1 going to a place of interest such as walmart or panera bread and same for cbg too and that’s how we’re getting our data next slide please and through that we have 649 cbgs which are the rows and then over 15 000 places of interest which are the columns walmart nera gym etc and then that data ranges from january 2018 to june 2021 and we have data on a weekly basis and through that we have over a billion data points to work with and the workflow in this regard would be just pre-processing the data a little bit uh doing some matrix or just creating the matrix then creating the sparse matrix and now getting to matrix factorization non-negative and regular this is just to cancel the noise get rid of the outliers and keep the signal then we look at our approaches create some models see if the models are scalable sample them and then just fill in the gaps with more factorization just so it’s more scalable next slide please and these are some of the approaches we’re looking at um approach one is just looking at the previous week’s data and then just using that to predict the next week approach two is looking at the previous four weeks of data to predict the next week model three is to look at the previous four weeks on a weighted scale to predict the next week then we also tried some regression models and we also tried long short term memory lstm given that the approaches we have tried these are the results we have we have tried all of the approaches on the raw matrix or the actual matrix and also on the factorized matrix and we used tucker for the factorization the models we tried with like model one two three regression polynomial regression with the dedgree of 2 and lstm and as you can clearly see if you focus on the right side’s plot you can see that the model two and three is outperforming all other models the regressions we have and the lstm and for that reason with a smaller subset we would only focus on model one two and three basically the baseline approaches to evaluate and if we go to the next slide as you can see here we have the subset this is not a geographical subset but a subset in terms of the number of poi since we are selecting the enter fairfax county but only 1670 poi and these are the most visited pois and we are focusing on the first 52 weeks which is basically the year 2018. and these are the results from 2018 and if you compare the rmse score between model one two and three you may see that the factorized matrix always tends to perform better than the raw or the actual matrix and this has been observed in all cases and model two and three performs better than model one even though in this case model one tends to perform as good as model two and three but we suggest two or three for the further approaches and going forward we you’d like to focus on the cbg and poi key trends and we’d like to develop maybe a multi-threaded approach for a larger area and we are in the process of trying some of the sophisticated approaches and hopefully in the next iteration of our work you will be able to see that we’d like to give a special thanks to the national science foundation the aspiring scientists summer internship program and the summer team impact projects