During this times of the global Covid-19 pandemic when pharmaceutical companies are scrambling to get the vaccine out, I think I should share my two cents on computer aided drug discovery.
There is no substitute to clinical trials and the classical case of developing a vaccine takes around 10 - 12 years but the current situation is different. This warrants open clinical trials.
What is a medical drug ?
It is a chemical substance which blocks or activates a certain protein. Virus, simply put is a speck of protein.
In my startup days, we developed something, we called contextual search, which simply put, is suppose you search for Sachin Tendulkar the search engine automatically popped up with a suggestion, say, Rahul Dravid.
How we achieved it ?
We had downloaded the entire Wikipedia data and linked the associated names using category. If you scroll down to the bottom of any page, you will find associated categories with the entity on whom the page is written; for e.g. "Cricketers at the 1992 World Cup" for Sachin Tendulkar and so on. These pieces of information was used to relate other entities falling in the same category ordered by the maximum matches.
How can it be applied to drug discovery ?
If one looks at the economics behind it, the more accurate the computer prediction, the lesser the time to bring it to market because of the likelihood of success in clinical trials.
Generally pharma companies start with a pretty huge list of chemicals whose physical, electromagnetic and other properties are already known. The next stage is identifying the similar properties of the harmful bacteria or in the present case the virus. Contextual search can kick in at this precise point. In the parlance of pharma companies it is known as virtual screening.
The three main methods are as follows.
1. Molecular Docking
2. Quantitative Structure Activity Screening
3. Pharmacopeia Mapping
The first method does not need any explanation whereas the second is matching the physical properties of the foreign body to the possible drug and the third is probably about the electromagnetic properties.
Thereafter one has to run a slightly modified form of contextual search (or one can call it an advanced search) where certain categories are also matched along with the entity. Basically if one is using a open source search engine like
Solr one has to match on multiple fields (in this case such properties) from the pre-existing database of chemicals.
There is no doubt that the above is an over-simplified explanation of a very complex process but my fellow computer engineers can get a basic idea from a fellow novice.
