Hunting down Covid-19

Image Source: Meseum fur Naturkunde Leibniz

Testing for COVID-19 is a search problem. It is like a maze. We are chasing the virus COVID-19.   Maybe we can use the model from the search algorithms that artificial intelligence is using.   There are several algorithms that can assist us to model our hunt for COVID-19.  But the issue is to find the optimal solution that will save time, effort and money to hunt down persons affected by COVID-19.   One solution that can assist us to hunt  COVID-19   is what data scientists call the depth-first search algorithm.  This algorithm follows a pattern of exploration. It explores in the order last come first.  The last person is first explored.  This search algorithmic model is profoundly effective when there at the first stage of the epidemic.  When there is no community transmission and we have sealed our borders effectively.  

Our hunt for COVID -19   is almost only based on a symptomatic approach.  We are only testing people who exhibit symptoms.  But this approach leaves out those who are asymptomatic.  To bring the asymptomatic persons into our exploration, we will have to test persons who exhibit symptoms of COVID-19, as well as all persons who have entered Goa from outside,  from a pre-established time frame. This means we set a certain time frame that aligns with the potency of a place being infected by the virus and do the testing of every individual. This is a set that is closed at both ends.   This exploration has to begin with the last person first till we explore all persons who have entered Goa from abroad and reach the first person who is at the beginning of time that is marked for exploration.  This depth-first approach for the hunt of COVID-19 is good but may not be optimal as it requires a complete sealing of the borders and total lockdown.  

Maybe there is still another approach that will be more useful. The search algorithm that is called the breadth-first search may provide us with ways of testing alongside normal hunting of symptomatic patients for COVID-19. This approach does not require the sealing of boundaries as it orders the exploration in a pattern first come first out.  It is like a set that is open at one end. This approach will test the first person who came into Goa from the time that is considered COVID-19  prone zone and reaches to our day.  This approach does not require sealing of our borders as we can keep testing those who enter.  Having considered both approaches, we find that the breadth-first approach is more optimal.

 There are other search algorithms that we can use as models for our growing pandemic. There is a greedy-best first search algorithm. It is a heuristic method.  It becomes more optimal as it is based on a model that does not look for exploration of every square of the maze but takes a coordinate that is closest to the goal.   This is calculated by measuring what is called the Manhattan distance. Thus greedy-best search algorithm minimizes the tests by chasing COVID-19 in clearly demarked hotspots and explore for COVID-19  in every person within the hotspot.  In India, we seem to be following this approach. Though this approach is optimal, it is not full proof.  There can be several linkages through which the virus may afflict our people.  But saves money, labour and is better than just symptoms alone hunt of COVID-19. Goa does not follow this approach as we have not seen a rise of infection as well as noticed its rise in a territorially bound zone. 

We in Goa need to follow this inspiration from the algorithms to deal with the return of the seafarers we may face potential dangers of seeing more infections.  To handle this scenario we will have to begin with strict quarantining of everyone that is brought into Goa. Besides quarantining each of them, we have to cover for the asymptomatic vectors amidst them. This will mean, we will have to test each of them.  Now we are dealing with a maze closed at least from one end.  The other end can remain open and keep adding new arrivals. Since we have to test each of them, we can follow the breadth-first approach that will test in the order first come first out arrival. It all depends on our testing capacity. We may not be able to test thousands in one go hence, first come first out the method which stays open to the addition of more individuals to the set while testing is going on appears to be more effective. 

We can also use another algorithm that is called A* ( A star) as our guide. It can bring more optimality to our hunt for COVID-19. This means from the set of the quarantine individuals, we will have to first take up those who have arrived from COVID-19  infected areas and hence are more likely to be affected. Along with those that exhibit symptoms to cover for the asymptomatic vectors of COVID-19, an optimal approach will first test those who have come from COVID-19 infected areas and therefore more likely to be vectors.  Again this approach is not full proof but is optimal as it will save time, money and labour.   The optimal solution is a mainly cost-effective solution. But against a lethal enemy like COVID-19,  it may not be really optimal as there can be linkages that we have not accounted for while we hunt the virus with our concern to keep our costs low. 

 Maybe we have to factor in the agency of COVID-19  in our hunt.  Maybe we can think of a game as a metaphor and not simply a maze to hunt down COVID-19 who is also an active player on the other side. Therefore, things are a bit more complex. Do we still have any model from artificial intelligence that can assist us to chase this virus more effectively by also anticipating the moves that it can make?   The model of the game becomes a better tool to visualize our hunt for COVID-19. But there is a problem.  The virus is not reacting to our moves. It follows its own timeline for its metabolism and infects everyone in its path. It is not responding to nor reacting to our efforts of hunting it down. This is the best part of it. It is deterministic in character.  

Adversarial search algorithms run games that AI plays. It requires a responsive agent who plays after we have played our step.  This is why we may imagine a game as a better metaphor to portray our hunt for COVID-19 but we cannot use the algorithms of adversarial search problems that are linked with AI playing games like chess, AlphaGo, etc.    At the most we can imagine that like in a game, we are closing the multiple options that are open to COVID-19.  As we block these options that are available to COVID-19 through testing, lockdown, hand hygiene, use of masks, sanitization of the work area, sanitization of persons through sensitization channel and social distancing.  We have observed all these steps to blocks all options available to COVID-19. Only then we can win this game. 


Is artificial intelligence good news, bad news or simply fake news?  It is certainly good news but has to be used within limits to honour human dignity, freedom, and privacy.  In times of pandemic, it is a great resource.  In fact, it can help us to race against the clock to hunt covid-19 from spreading like a wildfire. Besides catching up with the metabolism of the virus, AI may assist us to see the pandemic before it actually explodes. It means AI could see a pandemic a mile away.  In fact, Bluedot, a Canadian firm that specializes in infectious disease on 31st Dec. 2019 had already forewarned the world about this coronavirus pandemic using AI. It warned its clients from travelling to danger zones like Wuhan much before the Government could issue travel advisories to their respective citizens. AI cannot replace expert epidemiologists but it can be a great tool in their hands to foresee an outbreak of an epidemic and track its growth.  We can follow all the stages of its growth. Data scientists identify four stages in the spread of a pandemic. They are the import of the disease, contact stage of the disease, community transmission, and global pandemic. 

AI can predict the spread of the virus over time and place.  For instance, we can put AI to predict how the virus will spread over a week in India.  To do this we need to use machine learning that needs data sets that will give us the status of the covid-19 pandemic in India and also similar trends of data sets of other countries at this time. These data sets provide us with data about a particular stage of the spread of infection. The government as well as medical officials think that we have not yet entered the stage of community transmission. So we have to access the state of the disease in other countries at the contact stage. 

 To do this we need to import some data libraries. The first library that we will need is pandas as we dealing with tabular data sets.  We need pandas to manipulate and analyse such data. We also need to import some visualisation libraries like matplotlib, seaborn, plotlyexpress, folium to represent or portray our results.  We then read/ filter the data sets by using the pandas with Python.  These data sets are official data given by the governments about the pandemic.  Reading of the data sets gives us the data frame.  We finally can highlight the data that we have generated by using a gradient background.  This will give us the spread of the disease in India in comparison to other countries if we have used an Indian data set.  We can do also   find similar data in our findings to compare the spread of the disease over different states in our country.  This exercise can give us the total number of active cases, the number of deaths as well as the number of persons who have been cured.  

So far we only have manifested the status of the disease in India. If we need to predict how it will unfold for a week, we have to do more.   Data analytics can assist us to do this. It can predict the coming future of this disease.  We need to study the trend of the rise of coronavirus in India as well as in other countries, say China, South Korea, Italy or the USA. We can read the data set to trace the daily progression of the disease in comparison to other countries say China, South Korea, Italy or the USA. Using the visualization library called plotly to represent the data frame we can study the status of the disease in India and compare it with other countries that we have included in our analysis.  The visualization provides us with two graphs:  one is about the rise of the disease in India and the other countries. Both are exponential graphs. We can compare the rise of the disease in each country with that of India.  If any of them shows a sigmoid curve, we can see how the disease is on a downward slope. The sigmoid is a bent curve that indicates that the pandemic is under our control.  China and South Korea that used AI solutions to hunt the covid-19 infected people and also treat them show a sigmoid curve.  We in India are not having the same resources as A I. This is why we are trying to control the rise of the disease to a pace that it can be handled by the medical infrastructure that we have. If we allow the disease to simply flare up wildly without undergoing the pain of lockdown or social distancing and hygiene, given our weak medical infrastructure, we will not be able to handle the situation.

We can also represent the comparison that we have produced country by country on a common canvas and study all the graphs of the country simultaneously.  At this point, given our data on the disease, we will not see that India is reaching a sigmoid graph. We have to contend with the fact that it is on a steady climb. We are slower than others but all the same, we are rising. Some say that the total numbers of cases are more than the reported cases in India. This is so because we are not testing enough. Only testing can keep us ahead of the pandemic. But given our medical resources, we can only chase the virus and not stand ahead of it at this stage. This means some people will die without being identified as Covid-19 patients and some may be asymptomatic vectors that seem to be rising by the day. 

To forecast how the disease will develop in India for a week.  We need to extend the frame of our analysis to include a worldwide data repository.  To do this we have to import data from the data repository uploaded by John Hopkins University.  And stay with python to analyse this worldwide data along with the data of India.  We need a data set called Prophet, an open-source tool developed by Facebook which can forecast and predict non-linear progressions as well as regressions. When we use prophet we only need to give two inputs. Date stamp (ds) and the values (Y) Y is a numeric measurement that we wish to forecast.   To generate a forecast of a week ahead we have to consider what is called interval width which indicates the confidence that considers the chance of us going wrong. This covers the uncertainty that is inevitable in predictions of these kinds.  This is how we can forecast how it will unfold with a tolerance limit. It can also predict the deaths as well as cures for a set time. In our case, the tolerance that we have chosen was a week. 

 Exponential growth over a long time tends to be more difficult to predict because the variables keep changing fast over time and error increase in prediction. This is why a week is a good tolerance limit for forecasting.  It has been calculated that 60% of the world population will have to be infected for the virus to show a decline in infections. The same is true about India. For now, we can only slow it down and without an effective medical cure,  we are destined to let the virus infect somewhere close to 60% of Indians.  It is frightening to go into the long term prediction of this disease.  We do not have herd immunity in our country and without any vaccine, things look profoundly doomed for us in India.  We need to unite and fight together to save all of us. We seem to be heading towards a point where we have to choose together between death or life for all.  


There are several applications of AI that can enable us to fight the global pandemic caused by COVID-19.  There is no magical solution coming from AI. China seems to have shown the way but at the cost of human liberty and privacy.  While we are trying to save the life we may allow a certain level of trampling of human freedom and privacy.  Privacy can be protected only by cryptographic solutions.  In a case of a lethal disease like COVID-19, the privacy rules have to bit somewhat different.  There is a need to find the right balance between being able to save the life of the people and the protection of their right to go unimpeded. This can only occur in free and open societies that have trusting communities that will enable mass participation. Societies that are based on fear of the other do not have trusting communities and hence may have fears about the use of these invasive technologies.  The medical community is already used AI in the medical context and the privacy of the patients has not been an issue that has come to the fore. 

 We in India have developed an App through which the authorities can track our movements of individuals as well as the individuals can receive alerts about zones that have high chances of passing on the infection to them.  But this app depends on us who are willing to use it and are willing to let Government keep a watch on our movements.  It is called AarogyaSetu mobile App. Its aim is to track the spread of coronavirus.  what is more important that one has to understand that these apps work effectively to a large extent but are not a hundred per cent full proof and are only clever estimations. AarogyaSetu App works only with other AarogyaSetu Apps. Those who do not have the App are not covered.  The App developers have publicly declared that the data will be destroyed in the Post-COVID-19 scenario.  Such a promise was also made by Cambridge Analytica but we know that they continued to use and sell the data commercially to their clients.  Hence, trust deficiency about the handling of the data even at a time of great human distress becomes a grave issue.  If this data is collected and put to use honestly only to deal with this pandemic, it can really assist us to stay ahead of SARS-COV-2.   Maybe the Government has to swear in the blood to keep the data of individual persons safe. 

Data surveillance can also be augmented by image recognition technology enabled by AI. It can track the movements of the persons that are infected as well as the people that he/she has come to interact with. This technology is also invasive and has raised the same issues of security of privacy. It is difficult to enforce mandatory tracking of this kind. What we need is voluntary dissemination of personal information.  China is said to have used face recognition along with temperature scanners to help contain the spread of coronavirus.  The technology was upgraded even to recognize faces that were wearing masks. Face recognition was also employed to check on those who break the quarantine period.  It was a tool to name and shame the offenders of lockdown.  It has raised several questions about privacy issues. But the Chinese Government retorted that anyone who has nothing to hide has nothing to fear.  But big data policing has its own ethical questions about the ownership and use of this data in the coming future.  AI can provide data but it is for the medical personal to make critical decisions to save people’s life. 

There some AI companies that have employed AI to generate Vaccines to deal with COVID-19.  AI is also used to try and generate a drug that will control and cure the infection of coronavirus.  IBM’s Watson is being used to identify a molecule that will protect the immune system of the human body and also fight viruses that invade our body.  Thus we can see that AI is being put to work on three important fronts that are therapeutically aimed to deal with COVID-19.  They are vaccine development, antibody development, and small molecule development.  Vaccination can prevent us from contracting the disease, antibodies can prevent the virus from replicating and small molecules can inhibit its replication.  

The small molecules are said to be the easiest and cheapest solution to the people already suffering COVID-19.   AI can take us into the world of computational drug design.  To get a small molecule to become an inhibitor in a coronavirus, it has to dock into it. Scientists already know the N3 inhibitor of the SARS coronavirus which has 80% genetic similarity with SARS-CoV2. Therefore, they know the docking or the binding site.  What they need is to generate a drug by using a drug database like zinc.  The effectiveness of a drug that AI designs is known after we have conducted trials.  

Insilico Medicine has announced six potential drugs that can work as inhibitors. They used deep learning in combination with rewording based learning to computationally design their drugs.  A group of Korean scientists also worked out a model that they call molecular transformer-drug target interaction (MT-DTI). Their result points out that atanazavir, the inhibiter of HIV would have the highest predicable binding ability. It seems that HIV inhibiter could potentially be also an effective treatment against coronavirus.  There are already eleven FDA approved inhibitors.  This would be very convenient as they do not have to pass the regulation hassles. But these findings are statistical predictions and we still need clinical trials of these drugs.   It has been also noticed that AI has identified existing drugs like remdesivir or valproic acid as possible drugs for the treatments of coronavirus.  AI can also help us discover a new drug that can potentially treat coronavirus. These drugs require approvals from regulatory authorities besides their clinical trial.  We cannot keep AI  out of our fight against coronavirus. 

Leave a Reply

Your email address will not be published. Required fields are marked *


Hypocrisy is the tribute that vice pays to virtue.

- Fr Victor Ferrao