Developing new drugs for COVID-19 using artificial neural networks?

In this blog post, I will share with you the ideas and goals behind the project 2203 “Neural networks in chemistry- Search for potential drugs for Covid-19“. As the title suggests, we will work on finding potential drugs for Covid-19 using artificial neural networks.

SARS-CoV-2

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) is a member of a large family of viruses called coronaviruses, first reported in Wuhan City, China, in December 2019. This virus had terminated all the activities around the world and had put a stop to everything for the past two years. It is the causative agent for COVID-19. Well, I believe there is not even a single human being who has not been affected, and in many cases (~582 million worldwide) infected by this virus. Although there are vaccines which provide great action against the virus, already developed by the pharmaceutical companies. There is still a risk of developing new mutations by the coronavirus producing even stronger variants on which the developed vaccines may not be able act, and hence fail to contain the spread of the future variants.

Development of vaccines

Now, when the spread of COVID-19 pandemic is reduced, all of us ask the same question, are we prepared for another pandemic? The best preparation against viruses is development of vaccines which can give long term protection to this generation and to the next generation. However, developing a vaccine is not easy, and require years of trials and tests of drugs, which are thought to potentially work as inhibitor on a disease or virus. Developing a drug is a long and cumbersome process where a big amount of capital and time is required.

Vaccines can help protect against certain diseases by imitating an infection. This type of imitation infection, helps teach the immune system how to fight off a future infection. While vaccines are the safest way to protect a person from a disease, no vaccine is perfect. It is possible to get a disease even when vaccinated, but the person is less likely to become seriously ill.

How vaccine development can be accelerated by using Artificial Intelligence (AI)?

Well, there are millions of drugs which may potentially save us from another pandemic. But the question is how to test such a large number of molecules. The answer to this question is modern computational tools such as artificial intelligence (AI), machine learning (ML), and neural networks (NNs). The development in advanced computational tools have connected many branches, and one such connection is among medical science, chemistry, and computer science. AI, NNs, and ML were constantly applied to predict the outbreak, and to calculate the molecular docking scores in search of COVID-19 inhibitors.

Artificial intelligence is a wide-branch of computer science that deals with the development of algorithms and techniques to solve complex problems which typically require human intelligence. Modern approaches in AI are focused more on machine learning (ML) and deep learning (DL). In simple words, ML is defined as the capability of a machine to identify patterns from the available databases, through a learning process. An advanced version of ML is DL which consist of developing algorithms to create artificial neural networks (ANNs) that can learn a process just like neurons of human brain.

ANNs is a group of artificial neurons that takes the available data as input, passes it through one or more hidden layers where a weight is applied to the inputs, and is directed through an activation function. The activation function performs the non-linear transformation on the inputs and send the predictions to the next layer, and the process is repeated for the total number of epochs. The predictions are compared with the outputs of the problem to see the goodness of the fit. Often, mean squared error is computed between the predictions and the outputs to check the efficiency of the training. The training procedure depends on a number of parameters namely, number of neurons, number of hidden layers, learning rate, activation function, optimizer, bias and weights.

Goals of the Project

The main aim of the project is to find potential drugs against the SARS-CoV-2 virus using Neural Networks and Molecule Descriptors. The protein responsible for the virus replication, 3CL pro SARS-CoV-2 (6WQF) is the targeted protein under this study. The Python libraries TensorFlow and Keras will be utilized to develop neural network models, and Dscribe library will describe the input molecules using its available molecular descriptors. We will look for correlations between the molecular structures of the investigated chemical compounds, and their computed docking scores from the neural network models. The potential drugs based on the docking scores will open new pathways towards the future research in search of potential drugs against coronaviruses and would help in preventing any future outbreaks.

Thank you for reading this blog post. Stay tuned for the final blog post which will contain some interesting results from our work, and who knows maybe we will be able to find a permanent vaccine against coronavirus. If you feel fascinated by the subject and would like to know more about the process of developing drugs, then read this article from a previous work. Feel free to ask any question or write your questions in the comment box below. See you in the final blog post!