Deepfake Molecules Accelerate the Development of Covid-19 Treatment

While the “synthetic media artificial reality”, which has come into our lives with the concept of deepfake for the last 3 years was perceived as a global cyber threat until the beginning of the year, it has unexpectedly become a life buoy in the different areas during the pandemic process. In the last article, we mentioned that deepfake has become an important wildcard for the cinema, TV series, entertainment and advertising industries to continue production even under quarantine conditions. In this way, protecting the connection of people with life can also helps prevent social isolation from damaging immunity. Deepfakes is such a powerful weapon that in the fight against Covid-19, not only does it connect people to life,    but the same weapon is used in the war to protect human life.

Scientists train models with the help of special algorithms through artificial neural networks and obtain hyper-realistic molecule designs in order to rapidly develop effective molecules for the treatment of New Type Coronavirus. Molecular development studies carried out with deepfake designs will not only contribute to the scientific struggle with Covid-19,  but will also play a role in the decoding of human microbiological codes, diagnosis and treatment of many other diseases.

For urgent treatment, it is necessary to accelarate inventions

Covid-19 New Type Coronavirus was also identified as 2019-nCoV and SARS-CoV-2 when it emerged on January 7 because it was a new species not encountered in humans before. The rapid spread of this deadly new coronavirus, which began to be known as Covid-19 led to a global emergency. This has revealed to the need to test a number of new biomedical technologies in real life. Because in the pandemic process, it was necessary to significantly reduce the time to develop new vaccines and drugs to fight the newly generating epidemics. It was crucial to bring new treatments and tests to a level that could conduct clinical studies on humans in as little as a week or two instead of a few years. As a matter of fact, the US-based biotechnology company  Gilead  took action at the end of January, when Covid-19 began to make its impact felt. It made an agreement with a hospital in Beijing to begin testing of Remdesivir on human, which is an antiviral drug, in Wuhan, where is the heart of the epidemic. American biotechnology company, Insilico Medicine, operating at the Johns Hopkins University Emerging Technology Center in Maryland also took action without delay. The company, defining its mission as “accelerating molecule discovery and drug development by developing new AI technologies” has launched a rapid study to identify molecules that could form the basis of an effective treatment against coronavirus. According to news of Fortune, it tooks just four days that Insilico’s ai-based system detects thousands of new molecules that could be turned into potential drugs against the virus. Insilico has published a library of new molecular structures that it has produced for researchers to use on the Internet. It also launched an initiative to synthesize and test the 100 most promising new molecule candidates.

This time, GANs produced realistic molecule designs

Insilico started to work on Covid-19 with the decision of the company’s founder and CEO Alex Zhavoronkov on January 28. The company focused on an enzyme called protease that is critical for the reproduction of New Type Coronavirus. Because this enzyme was similar to “other virus matching models like Covid-19” developed by the famous researcher Assoc. Dr. Zihe Rao. from Shanghai Technical University.

As of January 31, Insilico has begun using 28 different machine learning models to design new small molecules that could bind to this enzyme and inhibit its function. Generative Adversarial Networks (GANs), which are used to generate deepfakes are included in these techniques. In this study, GANs produced images of new molecule structures of the correct quality to bind with protease, rather than fake videos that look very real. After four days-study, Insilico has produced hundreds of thousands of new molecule designs and published a library of thousands of potential molecule design options that meet the criteria for vaccine development work. Insilico has also published an article detailing free and unprecedented research on useful molecule design.  It invited in the researchers to examine and criticize the molecule designs, it developed in hopes of speeding up the process to develop a coronavirus treatment.

Insilico is not alone in its search for AI-based treatment

Insilico was not the only enterprise that hoped ai would help develop new treatments for the New Type of Coronavirus. A 3-person research team from Mathematics, Biochemistry and Molecular Biology, Electronics and Computer Engineering departments of Michigan State University published an article on this subject on January 31. The article titled “Machine Intelligence in the Design of Covid-19 Drugs” evaluated the use of ai-based Generetive Network Complex (GNC) in this area. The AI Cures research group at the Massachusetts Institute of Technology (MIT) also called themselves “a group of machine learning and life science researchers who developed deep learning methods to find promising antiviral molecules for  Covid-19 and other emerging pathogens”. The group draws attention to the “acute need to develop fast and effective therapeutics against the Covid-19 Pandemic, pathogens and health threats”. The group, arguing that traditional approaches to drug development are expensive and too slow to react to epidemics like Covid-19” emphasizes that AI tools have the potential to accelerate this effort.  AI Cures has made available a series of data sets in open source form to researchers to find effective antibodies against secondary infections that lead to the death of the patients, having poor immune system. The group is asking researchers to share their ai-based modeling to help find treatments that could possibly save millions of lives.

If the secret of proteins is cracked with 3D modeling …

DeepMind, which is another ai initiative also took an important step in 2018, based on the vision that ai research can accelerate new scientific discoveries. DeepMind threw together the experts in the fields of biology, physics and machine learning to apply the latest techniques to determine the genetic 3D structure of a protein.  AlphaFold, the use of large genomic data algorithm developed by the DeepMind team to determine the structure of proteins has advanced significantly in overcoming the basic challenges in biology with its very realistic 3D protein models.

Proteins are large, complex molecules that are essential to life and are found in all cells. Every function of the body (moving muscles, sensing light, converting food into energy, etc.) is associated with one or more proteins. That’s why it’s important to watch how they move or change. Descriptions of proteins called genes are encoded in human’s DNA.

What any protein can do depends on its unique 3D structure. For example, the antibody proteins that make up our immune systems are ‘Y-shaped’ and look like hooks. Antibody proteins that lock onto viruses and bacteria detect and label disease-causing microorganisms for destruction. But understanding the 3D shape of a protein from its genetic sequence using experimental techniques has been a challenging, complex and costly task for scientists over the past 50 years. The larger the protein, the more difficult it was to model.  Because there were too many interactions between amino acids to consider. It could take longer than the age of the universe to enumerate all possible configurations of a typical protein before reaching the correct 3D structure.

It is of great importance for scientists to determine the shape of a protein correctly. Because the diagnosis and treatment of diseases believed to be caused by misfolded proteins such as Alzheimer’s, Parkinson’s, Cystic Fibrosis seems to depend on this. As more information is gained about how proteins work through their images and simulations, opportunities for drug development will increase and costs will decrease. Consequently, millions of patients around the world will have more hope for cure.

AlphaFold Algorithm produces deepfake of Covid-19 protein structures

Upon the rapid increase in COVID-19 cases, the AlphaFold algorithm, which models DeepMind’s protein structures is began to be used in this area. Typically, producing a three-dimensional image from an amino acid sequence requires lengthy and costly laboratory work with a wide variety of protein visualization techniques. However, AlphaFold can shape a three-dimensional protein structure in the process of Covid-19. AlphaFold, won the CASP13 Competition (Critical Evaluation of Techniques for Protein Structure Prediction), which is held every two years around the world overcomes this challenge with artificial neural networks that determine the distances and angles between the amino acids scored by gradient descent. 

The Francis Crick Institute is also conducting research on the biology of New Type Coronavirus and how it interacts with cells in the body. A cross-verification of the structure of the spike protein determined experimentally by the Institute was carried out with the DeepMind algorithm. AlphaFold also created three-dimensional images for proteins whose structures are not easily determined. These protein structures were intended to help future drug development for Covid-19, potentially related to new drugs or therapeutics.

Perhaps more people do not know the meaning of deepfakes, but it is not easy to measure. On the other hand, those who perceive it badly are more than ones, who perceive positively . Those who abuse science and technology have led to so much damages over the years and this our anxiety has moved ahead of our trust.

We hope that the new molecules, whose designs are revealed faster and more effectively with deepfake technologies will make important contributions to the treatment of Covid-19. Thus, we are awakened from both the pandemic and the deepfake nightmare at the same time.