Data analysis on relationships between pharmacologic substances or targets & disease

Description of data

The analysis is dependant to two datasets that contain valuable information about the relationship of several chemicals & drugs with coronavirus.

• In the 1st dataset these relations are directly reported.

• In the 2nd dataset the pharmacologic substances show their effectiveness on coronavirus through genes, proteins or receptors.

• Also, we have available data from ATC, a system that classifies active ingredients of drugs according to the organ or system on which they act.

Description of the analysis

The analysis is divided in 2 cases regarding to which dataset will be used.

1. Based on the 1st dataset:

• First distinguish drugs from chemicals based on ATC’s list of drugs.

• Second find the Top 10 Key Opinion Leaders for every drug identified in our data.

2. Based on the 2nd dataset:

• First identify the drugs as before and for each drug to report the corresponding collection of proteins and coronavirus concepts.

• Then show which substances have similar targets.

Installation & run project

These are the packages used in the analysis:

library(stringdist)
library(tidyverse)
library(stringi)
library(readxl)
library(writexl)
library(quanteda)
library(igraph)
library(ggraph)

git clone https://github.com/askandali/coronavirus-analysis.git
cd coronavirus-analysis

install.packages("stringdist","tidyverse","stringi","readxl","writexl","quanteda","igraph","ggraph")
Rscript coronavirus_analysis.R

Drug Identification

In order to identify which substances of our list are included also in ATC ontology, we created a mechanism that checks for all of our substances :

• If the ATC list has a drug with the same name or part of it (e.g. zinc ~ zinc products).
• Or if there is a drug with similar name (e.g. cyclosporine ~ ciclosporin). This was implemented by calculating pairwise string similarity scores.

Identified drugs in the 1st case & text visualization based on the frequence a drug is detected

The full list of identified dugs is in the output folder with name “IIdentified_Drugs_case_1.xlsx” file.

The word cloud consists of all the substances that were identified in ATC and the size corresponds to the frequency of this substance in our dataset.

Top 10 Key Opinion Leaders per Drug

A key opinion leader (KOL) is generally an experienced physician trusted to give candid feedback to pharmaceutical suppliers. They are also knowledgeable about industry news, clinical trials, and other developments relevant to their specialties.

The full list of key opinion leaders per drug is in the output folder with name “Top10_KOL.xlsx” file.

Proteins & Coronavirus concepts per drug

The identification of the drugs from ATC list was completed for the case 2 dataset, by following the same identification technique.

The full list of identified dugs is in the output folder with name “IIdentified_Drugs_case_2.xlsx” file.

As part of the analysis, a classification over proteins and coronavirus concepts was implemented:

The full list of identified dugs is in the output folder with name “Proteins & Consepts_per_Substance.xlsx” file.

Clustering based on protein target

Additional information

• The identification algorithm in some cases found a drug with several names in ATC. For example, “vaccines” was found in “Meningococcal vaccines”, “Rubella vaccines” and more.

• The Key Opinion Leaders (KOL) were found by extracting all the scientists/authors related to the specific drug. Then, by calculating how many times someone has been referred to this particular drug, we were able to report the Top 10 KOL.

• Some drugs have less than 10 KOL, because less than 10 scientists/authors have been referred to them. For instance “riboflavin” has only Bowen Richard, Marschner Susanne & Keil Shawn D.