Using AI to discover new drugs for COVID
Developing new drugs to act against COVID-19 is a current major necessity. Yet, before launching a new drug product into the market there are many steps that need to carefully followed, since the pharmaceutical industry is one of the most regulated industries in the world. First, a drug candidate needs to be discovered. Then, pre-clinical tests need to be carried out to evaluate the toxicity of the compound and to access its pharmacodynamics and pharmacokinetics. The next step is the clinical tests, in which the drug candidate needs to be tested in humans. Finally, the product needs to be launched in the market. All these steps can take up to 15 years.
A huge effort in the drug development pipeline goes in the pursuit for new drug candidates. In the past, this was done using an irrational approach by trial and errors, i.e., many substances are screened based on empirical observation. Today, however, the pharmaceutical industry takes a rational approach, in which the drugs are designed to interact with the target of interest. To speed up the process and reduce the resources needed, computational techniques are being used to help engineering new drugs.
Among these computational tools, Machine Learning (ML) is getting much attention. ML is a technique in which the computer learns on previous experience. Based on the patterns found in the data from the system of interest, the algorithm creates models. These models can be used, for instance, to predict physicochemical properties of new molecules. They can also be used to predict drug-target interactions (DTI). Predicting DTIs is something that is of major interest in drug discovery, since it can be used to screen drug candidates.
ML encompasses many different concepts, among them Neural Networks (NNs). NNs are based on the architecture of the brain and its neurons. In simple words, each neuron represents a mathematical function, called activation function. The neurons are grouped by layers, in which the output of all the neurons of a given layer goes to each of the neurons of the following layer.
Due to the COVID-19 pandemic outbreak on December 2019, the development of vaccines and medicines to be used against this disease ramp up rapidly. Many efforts have been made to develop a vaccine to protect the people from the SARS-CoV-2 virus. Yet, developing a medicine is also of great interest, as it can be used as treatment for people that already contracted the disease.
In a previous work, Marián Gall and Michal Pitonák, the mentors of this project, used Neural Networks to predict docking scores of chemical compounds. The docking score is related to the biding energy between a chemical compound and a target, and is, therefore, a measure of the DTI.
In our project we are also using AI to develop models to help the search for such drug candidates. These models are used to predict the interaction between a drug candidate and one selected target from the SARS-CoV-2 virus. This target is the 3CL protease, which is a protein related to the replication of the virus. Giving what has been done, the goal of the present project is to further develop the work of our mentors. The intention is to continue predicting the docking scores of chemical compounds, but also to develop new models that can have a better accuracy.
One of the challenges in dealing with computer techniques to model the properties of molecules is to represent them. Although there are different standard representations for chemical compounds, such as the IUPAC nomenclature, they cannot be used in ML techniques. This is because the molecules need to be represented in a way that the computer can interpret them, since it only works with numbers. Furthermore, the representation should be in a form that can be used as an input for NNs.
There are different techniques that converts the molecules from the form that we know into a representation that a computer can work on. These representations are also called descriptors. Example of descriptors are Coulomb Matrix, Atom-centered Symmetry Functions, Smooth Overlap of Atomic Positions, Many-body Tensor Representation, and many others. In their work, our mentors used the Smooth Overlap of Atomic Positions (SOAP).
In this project, however, we will explore other descriptors that are available. The new descriptors selected were the Coulomb Matrix, the Atom-centered Symmetry Functions, and the Many-body Tensor Representation. After obtaining the models using these other descriptors, the goal is to evaluate if one of them will result in better models than the SOAP.
Hii Gabriel , that’s a really good post !
Such an interesting and current theme!
I really liked the images for better visualization 🙂
Good work for the next steps!
That is some great and crucial research! Creating a reliable model is always hard, and the use of a neural network is indeed a great method to improve the modeling! Wish you success in this project
Amazing job!
An astonishing project with a crucial and pertinent purpose involving the pharmaceutical industry and thus the public health. Furthermore, the explanation about the topic was concise and intelligible. Great work!
Great project! Such a challenge! Artificial Intelligence will accelerate the launching of new drugs into the market for covid treatment and many other diseases.
That’ is a great reserarch.
Good luck! The use of AI will probably accelerate relevant findings. Let’s win the fight against this virus!
Ótimo tema! Sucesso!
That’s a really interesting and important research.
Which tool will you use to these descriptors? (Coulomb Matrix, the Atom-centered Symmetry Functions, and the Many-body Tensor Representation)
And which tool will you use to program the NN?
Excelente texto. Sucesso. O projeto.
Congrats for sesrching new therapy and drusos.
Excelente projeto . Parabéns pela pesquisa , que tenham ótimo resultados .
Congratulations for your research. Very pertinent.
Parabéns, Gabriel 👏🏻👏🏻
Excelente trabalho!
Parabéns
Parabéns pelo trabalho!!!
Sucesso em todo projeto.
A key aspect of smarter A.I in general is that it will open up new opportunities for increasingly capable humans and machines to collaborate more closely, leading people to a better quality of life.
Muito bom….excelente texto e pesquisa.
Hi, Gabriel! It seems a very interesting research. I would like to know which kind of read data are used to train NN’s
Amazing research project! Using AI to reduce time and cost in drug development will be the future in this area. Looking foward to see the results.
Very interesting and well written! Excited to learn more about this project in the next posts.
Amazing theme.
Congratulations
Such an amazing and important research topic within the context we live in today! All the best luck for the next steps and already fondly anticipating the results!!
Excelente projeto de pesquisa e conteúdo.
Parabéns!
Awesome!!’
Excellent theme! Health treated with respect. This articule reports that new drugs take up 15 years to be developed and approved. It is hard work and must be done for benefit of humanity, not financial.
Very interesting using AI to develop a new drug. Great project!
Very interesting research!
Trabalho excelente. Parabéns, Gabriel!
Great project!
Excellent
Great project, well done
Wow! Very interesting project.
Excelente, usando a tecnologia e a AI para benefício em massa! Sucesso!