Nebula Graph is an open-source distributed graph database: https://github.com/vesoft-inc/nebula
This post will detect the Coronavirus spreading based on a real case that took place in Tianjin, China. In this case, there are five confirmed cases of the nCOV pneumonia in the same shopping mall in Tianjin. From the initial three cases, it seems that there is no epidemiological correlation. Against such background, how to uncover the links among the cases?
Evidences have shown the nCOV transmitted from person to person. I.e., if extracted the transmission in graph model, a person transmitted to one another through an edge (Demo 1). Consider that A infects B, then B infects C, then C to D… This makes the tree-like path (Demo 2). However, given cross-infection, repeated use of the public places and transportation, the spreading path of the virus becomes a network structure.
Thus a graph database is the best choice to store and explore the transmission relations. In this post, we will discuss how the nCOV disease spreads and who are the possible suspected cases.
Let us use Usr1, Usr2, Usr3, Usr4, Usr5 to refer to these five cases, and look at their tracks:
Based on the above introduced data, we extract a data model with two vertex types, i.e.
, one edge, i.e.
1. ID: unique identification of a person
3. ConfirmedTime: used to trace the order of the patients’ onset
Based on the above model (the figure below), we can import data. Then with the help of Nebula Graph, we can find out the source of the virus, and who should be observed / isolated after the diagnosis of a patient.
Import the above data into Nebula Graph to build relationships among persons and places. Take Usr1 as example:
-- Insert Usr1 INSERT VERTEX person(ID, HealthStatus, ConfirmedTime) VALUES 1:(2020020201, ‘Sick’, '2020-01-24'); -- Insert place INSERT VERTEX place(name) VALUES 101:("Shopping mall Area A") -- Insert edge INSERT EDGE stay (start_time, end_time) VALUES 1 -> 101: ('2020-01-23 12:00:00'， '2020-01-23 18:00:00') -- Insert another place INSERT VERTEX place(name) VALUES 102:("Community A in Hepin District") -- Insert another edge INSERT EDGE stay (start_time, end_time) VALUES 1 -> 102: ('2020-01-23 18:00:00'， '2020-01-24 8:00:00')
Together, let’s uncover the mystery of Usr1 infection step by step.
1. Find out where Usr1 was on January 23
$PlaceUsr1Goto = GO FROM 1 OVER stay WHERE stay.start_time > '2020-01-23 15:00:00' AND stay.start_time < '2020-01-23 23:00:00' YIELD stay._dst AS placeid
2. Check if Usr1 exposed to any confirmed cases during this time
GO FROM $PlaceUsr1Goto OVER stay REVERSELY WHERE $$.person.HealthStatus == 'Sick' AND $$.person.ConfirmedTime <= "2020-01-23"
It is strange that at the time of Usr1’s onset (2020-01-24), there was no fever in the people he contacted. Could it be that these people have come into contact with other patients (thus becoming carriers)? Let us continue our analysis.
3. Check who have an undirected connection with Usr1
$PersonUsr1Meet = GO FROM $PlaceUsr1Goto OVER stay REVERSELY YIELD stay._dst AS id $PlaceThosePersonGoto = GO FROM $PersonUsr1meet.id OVER stay YIELD stay.start_time AS start stay.end_time AS end GO FROM $PlaceThosePersonGoto.id FROM stay REVERSELY WHERE $$.person.HealthStatus == 'Sick' AND $$.person.ConfirmedTime <= "2020-01-23" -- become sick before this time stay.start_time > $PlaceTHosePersonGoto.start AND stay.end_time < $PlaceThosePersonGoto.end -- have connected
We found that Usr1 had connection with Usr2, Usr5 between 12:00 on January 23 and 8:00 on January 24. Both of the two were healthy at that time. But Usr5 had previously contacted the patient Usr4 with a fever.
So far, we have found the spreading path:
Usr4 became sick on January 21. After being sick, he went to a hot pot restaurant in Nankai District, Tianjin (11:00-20:00 on January 23). Here he was exposed to (then healthy) Usr5 (11-15 pm on January 23), making Usr5 a carrier during contact. Then Usr5 headed to Tianjin shopping mall A, B, and C area (16:00-23:00 on January 23). During this time, he transmitted the virus to Usr1 who worked in area A (12:00-18:00 on January 23). And Usr1 became sick on January 24.
4. Find out who needs to be isolated
After Usr1 is diagnosed, we need to see where and when she has been to and who was in contact with her in the same place during this time period. People that were exposed to her need close observation and isolation.
GO FROM 1 OVER stay YIELD stay.start_time AS usr1_start, stay.end_time AS usr1_end, stay._dst AS placeid | GO FROM $placeid OVER stay REVERSELY WHERE stay.start_time > usr1_start AND stay.start_time < usr1_end YIELD $$.person.ID
We can see that Usr1 and Usr2 have connected with each other in a Community A in Heping District, Tianjin, which made Usr2 a suspicious case.
The following figure shows the visualization of the above analysis.
Of course, if you want to observe large amount of vertices, such as tens of millions of potential people and their second and third propagation trajectories, a program with batch queries will be more efficient.
The Spring Festival travel rush and other causes lead to the wide spreading of the nCov. We noted from social media that all the communities, villages, businesses are adopting extremely stringent quarantine and asking people to report daily whereabouts and health status. Both the quarantine and track of billion people request huge resources in time and money.
But such self-report mechanism is inefficient and unreliable, especially there are always cases of concealing behavior and medical history. This may lead to failure in timely isolation and treatment, and also impose negative affect on business.
Fortunately, the development of big data has facilitated the construction of the data system in security, transportation, medical departments. In the above Tianjin case, we used a few cases to demonstrate how graph database helps to locate suspicious cases and decrease the risk of infection.
Previously published at https://nebula-graph.io/posts/detect-corona-virus-spreading-with-graph-database/
Create your free account to unlock your custom reading experience.