New contributions were added:

Detecting Global Community Structure in a COVID-19 Activity Correlation Network by Hiroki Sayama
https://community.wolfram.com/groups/-/m/t/3056172

COVID-19 SIR models: transmission, vaccination, herd immunity dynamics revealed by Athanasios Paraskevopoulos
https://community.wolfram.com/groups/-/m/t/3008488

Data modeling by Thomas Colignatus was added:

Redesign of didactics of S(E)IR(D) -> SI(EY)A(CD) models of epidemics
https://community.wolfram.com/groups/-/m/t/2004784

New app for epidemiology education from Rui Alves was added:

EpiPlay: using Mathematica to gamify education in epidemiology
https://community.wolfram.com/groups/-/m/t/2535927

New data analysis from Peter Riley was added:

Deep neural network detection & clinical staging of COVID-19 chest X-rays
https://community.wolfram.com/groups/-/m/t/2389110

New vaccine data analysis from Damian Calin was added:

Covid-19 vaccine campaigns efficacy analysis
https://community.wolfram.com/groups/-/m/t/2383314

New data analysis from Arshaan Sayed was added:

Correlating COVID-19 government measures to biweekly/daily outbreaks
https://community.wolfram.com/groups/-/m/t/2362327

New vaccine data analysis from Arshaan Sayed was added:

Analyzing COVID-19 vaccine sentiment over time
https://community.wolfram.com/groups/-/m/t/2317293

New resource from John Cassel and Daniel Lichtblau was added:

Finding and analyzing a COVID subvariant in Australia
https://community.wolfram.com/groups/-/m/t/2342489

New vaccination data visualization from Mads Bahrami was added:

CDC COVID19 vaccination data across US counties
https://community.wolfram.com/groups/-/m/t/2282418

New resource from Daniel Lichtblau was added:

Analyzing the spread of SARS-CoV-2 variants in Florida
https://community.wolfram.com/groups/-/m/t/2206874

New resource from Daniel Lichtblau was added:

Analyzing the spread of SARS-CoV-2 variants in California
https://community.wolfram.com/groups/-/m/t/2205357

A function to generate the most recent Google Mobility Reports for different countries: https://www.wolframcloud.com/obj/mohammadb/DeployedResources/Function/GoogleMobilityReport

A function generating a WL dataset from the OWID vaccination and testing dataset for the countries: https://resources.wolframcloud.com/FunctionRepository/resources/OWIDCOVID19Data/

New data processing resource from Mads Bahrami was added:

Mobility changes data: transforming to Wolfram Language dataset
https://community.wolfram.com/groups/-/m/t/2160386

A visualization resource by Isao Maruyama was added:

COVID19 Tokyo per days of the week
https://community.wolfram.com/groups/-/m/t/2133807

New visualization post from Mads Bahrami was added:

COVID19 data visualization across US counties
https://community.wolfram.com/groups/-/m/t/2119049

New post by Siria Sadeddin was added:

Face mask detection: classifying image data
https://community.wolfram.com/groups/-/m/t/2139499

A modeling resource by Diego Zviovich was added:

Epidemiological Model for repetitive rapid testing for COVID-19
https://community.wolfram.com/groups/-/m/t/2075883

New analysis from Seojin Yoon was added:

Analysis of the Change in Phillips Curve After COVID-19 with Regression
https://community.wolfram.com/groups/-/m/t/2055704

A resource by Luis Borgonovo was added:

A SEIRD Model For COVID-19 Using DDEs
https://community.wolfram.com/groups/-/m/t/1996374

A post by Kyle Keane was added. As an additional benefit, it includes valuable replies :

Ways to visualize COVID-19 simulation results?
https://community.wolfram.com/groups/-/m/t/1962739

I thought some of you might be interested in our recent work: https://arxiv.org/abs/2007.00159

We have tried to come up with an understanding of the precise influences of human demographics and settlement, as well as the dynamic factors of climate, susceptible depletion, and intervention, on the spread of localized epidemics.

We've also designed a dashboard for it (based on US counties): https://www.wolframcloud.com/obj/mohammadb/COVID19Dashboard2

Don't forget the upcoming free study group devoted to analyzing COVID-19 data. I look forward to joining you online on September 28, @Hamza Alsamraee. Here is the sign up link for anyone who is interested: https://wolfr.am/COVID19StudyGroup

While all the initial major outbreaks in the US were in major cities there have been lots of news stories about outbreaks in rural areas, especially around meat processing plants. So I wanted to see if the lower population areas in the US are catching up to the denser areas.

tldr, No, not yet at least.

https://community.wolfram.com/groups/-/m/t/1992898

Some simple visualizations:

100 Days of COVID19 Over US Counties

COVID-19 reopening criterion

Mads, This is great! Normalizing the Cases map by Population over counties for a given state would be interesting see. Sam Daniel, Tucson,

You can now get the COVID19 data at the US County level from WDR. Check out my short post on that.

The behavior on weekends is very interesting! It looks like people are doing their grocery and retail shopping on weekdays, and "sneak" out on weekends to the office.

I posted a simple model for assessing new caseloads in US cities.

https://community.wolfram.com/groups/-/m/t/1945196

This builds on work from a prior post that works with an estimate of the logarithmic derivative of tested-positive counts as a function of time. The extrapolation gives what amount to testable hypotheses in the sense that it predicts near-term caseloads (so in a few weeks I get to find out whether or not I completely missed the mark).

Should this model turn out to have some predictive power, an important virtue is that it is remarkably simple. In addition to using a only straightforward differential equation, it requires but few hypotheses and no fitting of parameters (all values come from existing data). Models this spare on underlying requirements are always good to have. When they work, that is.

In a very short essay, I’ve explored COVID19 fatality vs population age and income over US counties. My analysis suggests that the case-fatality ratio decreases by a factor of about 2 when the median household income goes from $30,000 to $60,000. However, it shows less sensitivity to incomes of $60,000 or more. Additionally, the case-fatality ratio increases slightly with the median age, as expected. On the other hand, if one considers the population of 65-yrs old and over, the case-fatality ratio shows some counterintuitive features, e.g., it decreases by this population, which is not what one expected (in fact, one may expect the reserve). The visualization of median house hold income vs population of 65yrs old and over suggests that the income increases by this population, which may imply better health care services and ultimately less fatality.

Are there data of deaths due to COVID-19 over time and age, separated by individuals' heath conditions, such as: 1. None 2. Asthma 3. Pulmonary 4. Heart and combinations thereof. Also, in view of Dr. Oz's observation that many thousands Lupus patients and others taking Hydroxychloquine, it would be very informative to determine their vulnerability to the virus collectively, or compartmentalized by other heath conditions as stated above. This granularity may add to a meaningful analysis and prognosis for a given population and enable improved measures to recommend activities at commensurate levels with respect to other associated imperatives.

In the post "An SEIR like model that fits the coronavirus infection data"

https://community.wolfram.com/groups/-/m/t/1888335

listed above under epidemic modeling, I discuss how to fit an SEIR-like model to the coronavirus data. I argue that there is value in doing this regardless of how good the data might be in that a good fit nevertheless allows one to understand the dynamics of an outbreak and make forecasts (how many cases, how long of an outbreak). The key to being able to find a good fit for the data is to estimate initial susceptibility correctly. I argue that the main effect of a lockdown is to lower this number drastically. In the discussions of the post I provide a thought experiment which illustrates why this is so. Susceptibility thus understood is also used to illustrate the effect of lifting containment restrictions too early. In the post there are models for several countries, including China, USA, Austria, Finland, France, Germany, Italy, Spain, and the UK. There is additional material, including a pdf document with smoothened curves of daily case tallies for several countries. I attach some pictures from the post. The models are updated on a daily basis. Please visit the post for the updates, they will not be reflected in this response. I attach a notebook with the simplest version of the model that needs to be updated but is fully functional.

Attachments:

ISEIRmodel6April.png

SEIR model C 27 ...nb

SEIRmodel6.png

SSEIRmodel6AprilWOR.png

USSEIRmodel5April.png

Here's an overview of some compartment based models used in epidemiological modeling: https://community.wolfram.com/groups/-/m/t/1920119

I was interested in this as well. Additionally thoughts on curve fitting to hospitalizations versus death rate.

The epidemiological models are useful for understanding different levels of dynamics, but it is hard to get all the rates right until the epidemic is over. Since coronavirus is assumed to be new to the human race, almost everybody should be susceptible. And since there is no treatment nor vaccination, then quarantine becomes the only effective method of control. The logistic model only requires that at some point quarantine becomes effective. At the outset, the infection can spread exponentially because almost everybody is susceptible, the exponential phase can go to complete diffusion because of easy international travel so a few hundreds of infected people can seed most of the world. Thus the only way to break the spread is effective quarantine. And the spread of quarantine methodology has to be faster than the spread of the infection. So with coronavirus now that there is effective testing to enhance quarantine, the quarantine method should halt the spread. Looking at the logistic model, early on in the epidemic there is little evidence that quarantine is working, until it fully kicks in, then control should be rapid. This is a graph I am finding useful. It is for the state of New York. The orange dots are the differences of the logarithms of the daily total cases. Thus it is a continuous growth rate. If growth were exponential it would be a horizontal line. And at the outset there is no pattern to the daily rates--they could be fit to any model, The blue dots are the course of daily rates predicted by the logistic model using all the orange dots for the fit. There is apparent convergence to the logistic model as the epidemic progresses. The default fitting method using Norm also favors the more recent points. This is a log plot of the cumulative cases in orange fit to the logistic model in blue. Since the squares of the later much higher values dominate, the early data are effectively ignored.

But in this case ignoring the early data is a good thing because early on there was not adequate testing and only the most obvious cases could be discovered. Since the counts are cumulative, cases which are eventually found end up in the cumulative totals, and it only takes three points with accurate information to fit the curve.

Since the quarantine effect can easily dominate, by spread of information almost instantly, quarantine could potentially break the epidemic in a maximum incubation period. For instance if a distance of six feet between each person for two weeks maximum incubation would break the transmission, then the epidemic could end in two weeks if everybody actually complied, and there were no people with carrier states.

I think the solution to the epidemiological modeling is to use the logistic model which only assumes quarantine first. Once the logistic model kicks in, it should give an accurate prediction of the susceptible population, as its limiting population, L. The rate parameter, k, is at the dashed line in the first graph, and this is likely a good estimate of the maximum exponential rate of spread for the virus. See Logistic Model for Quarantine Controlled Epidemics

But that model implies an odd method of quarantine. The population of South Korea is over 50 million. The model you show takes 17000 people, presumably infected and contacts, and puts them into a camp away from the 50 million population and then lets the infection run its course. Ideally a good quarantine model should keep initially healthy people at risk away from known infected people, with the goal that many of the people at risk will not become infected because of the quarantine. With effective quarantine new cases should stop within the few incubation periods it takes to trace the contacts.

@Robert Rimmer, nice observation! I think it fits very well with SIR model. So maybe SIR model can explains it

Please see notebook in Wolfram Cloud or attached below.

Attachments:

2021-06-04T14_38...nb

Any idea how many tests they are doing each day now? Do you know if this includes imported cases?

I added a post about US county-level timeline plots here: https://community.wolfram.com/groups/-/m/t/1918332

Thank you for gathering in one place all these COVID19 resources from Wolfram Community!

I've just added a new version of my map at Confirmed COVID-19 Cases in Catalonia.

Map of March's total cases with Tooltip info displayed for each municipality: www.wolframcloud.com/obj/bernate/covidcat

There's been so much great work from the Wolfram Community around this topic, and more information and data sets emerging every day than we can reasonably expect to digest within the company — I added a post at

https://wolfr.am/COVID-19-DATA

with some requests and recommendations for people who want to do more, whether it's just pointing out relevant data sources, or taking the time to make some of that data computable and more instantly ready for other Wolfram Language users to explore.

Hey, Stephen will be doing a live exploration of some COVID-19 data this afternoon (1:30pm CST, US) on his twitch channel: twitch.tv/stephen_wolfram and will be simulcast on the Wolfram Research Youtube Channel. Thought the folks on this thread may want to join :) Stay well

Hi all, Thank you so much for the insightful data and analysis!Impressive work! Could someone please share with me (notebook?) how the Ribbon and Surface model are created? ~ Best regards, Saar Hersonsky

Hey Brian,

Thank you so much for the information! I was able to Import both files with the full path you kindly provided in your reply:

Import["https://github.com/arnoudbuzing/wolfram-coronavirus/raw/master/data-files/QHD434151-putative-proteasecleaned-ribbon.stl"].

Could you kindly explain how you got the correct path above? I am not sure how the "raw" (just before the /data-files/) shows up in your message. It definitely does not show up when I use "copy path" appearing in my browser and therefore the Import did not work.

Best regards, Saar

Awesome!

I used your path and could not figure out why I failed...Indeed, I dowloaded and use Import on my local machine. Thanks so much and I will be back with more after I study the algorithms to produce these surfaces.

Saar

Another stream today! This one featuring Robert Nachbar discussing Epidemiological Models for Influenza and COVID-19. Will be livestreamed at 3pm EST on https://twitch.tv/wolfram

Hi, Yes, you can see them on our YouTube channel here: https://www.youtube.com/user/WolframResearch/videos With novel coronavirus specific videos here: https://wolfr.am/LaDVTtsu

I'm not really sure what you mean by Twitch videos disappearing? Twitch removes past broadcasts after 30 days, so we also upload the video to Twitch.

Please join us in our https://www.twitch.tv/wolfram discussion on Geo-spatial-temporal COVID-19 simulations and visualizations over USA next Tuesday March 17 5:30 PM EST. We'll take Anton's framework and apply it to create a model for the US.

Yesterday the John Hopkins data for the US was about 1000 cases too low but the last two digits 68 matched the worldometer numbers (I didn't track when it was corrected). There must be a lot of human intervention to produce the numbers so they can be misleading.

I am not talking about all of them having an exponential growth (I thought this would be obvious, but next time I'll spell it out :-) ) I am talking about the fluctuations in the last few days, which are common to the European countries in the plot, but not to Iran, the US or Korea.

If testing is currently limited by the manufacturing of kits, and most come from the same source, then you could be right.

Another suggestion I got is that the data reporting deadline (for this particular dataset) has changed, shifting some reported cases from March 12 to March 13. This seems the most plausible to me, so far.

Late reply. Probably you already figured it out.

There were no updates in those days for some countries. JHU have many issues in their datasets. https://github.com/CSSEGISandData/COVID-19/issues

caseData[[{17, 12, 201, 210, 405, 463, 32, 19, 21},    
{"Country/Region", "3/11/20", "3/12/20", "3/13/20"}]]

Here are a couple of other ideas. The case data for Italy is starting to converge, so you could use the L from that and estimate that the deaths stay a relatively stable ratio to cases. That might be a big assumption for Italy where the death rate seems too high. You can also track parameter convergence--this is dramatically convincing for South Korea. Code for these functions using case data for Italy and South Korea are in the attached files as well as code to fit to the differential equation, from the first derivative of the logistic function. Rather than using interpolation to get the derivatives you could manually draw approximate ideal slope lines to the data to get an estimate for k and L from the equation for the parabola. Also when the log plot starts to show downward concavity, use only the most recent points which will be in the logistic phase.

Attachments:

Italy.nb

South Korea.nb

Robert:

Thank you so much - that was a very helpful discussion, and I will go through the other discussion thread that you linked to.

The one problem with choosing to fit to an exponential model (because the outbreak is in its exponential phase), is that it eliminates L as a parameter, and I am interested in that value. Nevertheless, it's also a useful thing to have better short term estimates as you showed.

If I insist that my model also produces a value for L, I must accept that the estimate of this value will be poor while the outbreak is in its exponential phase. And that makes sense. I had hoped that a constraint might add some extra information to the model, but apparently this damages the parameter estimation. I assume L estimation gets better when the phase reaches its midpoint.

Thanks for your very helpful demonstration!

Steve Rector

HI.

In the attached notebook, I've fitted a Logistic model to Wolfram repository data for CoV deaths from Italy through March11, with 90% confidence bands out to March 15. Since we're still in an early phase of the outbreak, the bands diverge relatively rapidly as expected.

In the NLM fit, I've added a constraint to "L" based on the a-priori information that the number of deaths cannot be less than a number close to the present value. However, when I print out the ParameterConfidenceIntervalTable, the 90% CI for L is: {-3453.81, 15896.9}, with expected value 6221. I also get the warning: FittedModel::constr: The property values {ParameterConfidenceIntervalTable} assume an unconstrained model. The results for these properties may not be valid, particularly if the fitted parameters are near a constraint boundary.

Now, I expect there to be a very wide interval for L given the early phase of the outbreak. But, the Confidence Intervals are not taking into account my a-priori information, and the warning explains it. It seems to me there must be a method where the CI of the fit parameters can take into account this a-priori info (in my case, L > 1200). Is this a missing feature in Mathematica?

Stephen Rector

Attachments:

ConstrainedExample.nb

It is supposed to update automatically. If it does not, you can always delete it manually with

DeleteObject[ro]

Thank you @Hee-Young ! You perhaps would be interested to take a look at the work by @Yu-Sung Chang:

• DASHBOARD: http://corona19korea.com/index-en.html
• SOURCE: https://github.com/yusungchang/corona19korea.com

Dear Vitaliy and many other experts,

1. Is it possible to compile the data for the number of testing? So that we can get the ratio of the confirmed cases relative to those who get tested?
2. Is it also possible to redesign the data set to include the 'City', in addition to the current geographical classification, namely, country/region and administrative region? That way, perhaps we can get more detailed information about the containment and spread of the COVID-19 inside and outside of the city?

Best,

Thank you very much. I am also compiling some data for South Korea (where my parents are living). Once I am done, let me send you (share with other experts) the data set.

Dear Szabolcs,

I have personally compiled the various data for Korea, where you can find the estimated number of daily testing: https://github.com/interglobe07/COVID-19-Korean-Data (COVID19 Korean data Updates) I hope you can find it useful. Best,

I suspect the problem is early data collection. Iran probably does not have the resources to detect whether all exposed cases have become infected. Thus the cumulative case data will lag the actual cases until the backlog of cases in the community is discovered. The log plot of cumulative cases from the JHU data is still showing exponential growth.

When quarantine measures start to work or susceptible population significantly declines, the graph should start to show growth slower than exponential. Until that happens it won't be possible to predict the end of the epidemic.

In case it's not covered in data resources in OP, here is a history data source someone crawled from Ding Xiang Yuan (DXY), down to every cities of every provinces in China.

COVID-19/2019-nCoV Infection Data Realtime Crawler

https://github.com/BlankerL/DXY-COVID-19-Crawler/blob/master/README.en.md

Note the data source is non-official. DXY, as I know it, is an online non-gov society of doctors and nurses from mainland china. Their data could be different from officially published one.

I have studied the genetic sequences of COVID-19 and SARS-like viruses, using Chaos Game Representation and Z-curve methods (hyperlinks to my Community posts). Z-curves provide a fascinating visualization of genomes that helps a lot for classification and clustering. The hierarchical clustering of viruses identifies Bat coronavirus RaTG13 as the most-likely culprit of COVID-19. My results strongly support the hypothesis of a Bat origin of COVID-19. I appreciate any comment or feedback :-)

I just wanted to note for anyone who might be interested that the latest release of IGraph/M from a few days ago now exposes the igraph C library's SIR modelling functionality. It is fairly simple at the moment. It can run several simultaneous stochastic SIR simulations on a network, and only returns the S, I, R values at each timestep (not individual node states). It can be used to study the effect of network structure on the spreading.

http://szhorvat.net/mathematica/IGDocumentation/#epidemic-models

UPDATE: I just added another example to the documentation to clarify what this functionality is good for. If you've opened the above link before, please do a hard-refresh of the page (Shift-F5 on Linux/Windows or Command-Shift-R on Mac)

I have a new notebook titled 'china-province-graph.nb' here:

https://github.com/arnoudbuzing/wolfram-coronavirus/tree/master/notebooks

It contains the 'bordering provinces graph' (not a built-in dataset).

Might be useful with your IGraph package?

Hi @Per Møldrup-Dalum, I am glad you like our resources and we highly appreciate user feedback, thank you! For this specific type of question I recommend reaching out directly to our Wolfram Data Repository team at: https://datarepository.wolframcloud.com/contact-us Please note, Wolfram Data Repository entries are continuously updated and new information can appear on their pages in the future.

It would be neat to see a SEIR type analysis

I did a simple chart how 2019-nCoV aligns against SARS, MERS. Here results and source code.

https://gitlab.com/sparvu/mathematica

data={{24527,490},{8098,774},{2500,858},{1568,616}};
BarChart[data,
PlotTheme->"Business",
ChartLayout->"Stacked",
ChartLabels->{Placed[{"2019-nCoV","SARS","MERS","Avian Flu"},{{0.5,0},{0.8,1.2}},Rotate[#,(1.75/7) Pi]&],Placed[{"",""},Above]},
LabelingFunction->(Placed[Rotate[#,0 Pi],If[#1>1,Center,Above]]&),
ChartLegends->Placed [{"Infections","Fatalities"},Right],
PlotLabel->Style["2019-nCoV Infections",FontFamily->"Helvetica",Thin,24],
AspectRatio->0.55]

There is also raw data being collected here in the form of a Google Sheet. It relies on data abstracted by a human (a work study student at the University of Houston operating under my supervision) from the daily reports being produced by the World Health Organization. I attach a notebook that shows how the data can be sucked in from the Google Sheet and turned into a Wolfram Language Dataset. From there, I run a few basic queries.

Please see notebook in Wolfram Cloud or attached below.

Attachments:

2021-06-04T14_16...nb

• I have a few contributions here: https://github.com/arnoudbuzing/wolfram-coronavirus
• The notebooks from this repo can be viewed from this location: https://wolfr.am/JZNRriEE
• This notebook gallery was created with this tool: https://wolfr.am/githubviewer