|
The integration of AI (artificial intelligence) in pharma manufacturing and research is seen to reduce human error and speed up drug discovery, said Dr Alok Aggarwal, CEO and chief data scientist, Scry AI, a company developing enterprise applications using proprietary AI algorithms.
Discovering small molecules is particularly attractive because a large amount of high-quality data is now available in public and industry databases, and this can be used for training AI algorithms to provide more accurate predictions. Moreover, chemical structures and properties of smaller molecules are described more easily and are more complete, and researchers have a better understanding of the underlying interactions and potential toxicity, he added.
The traditional process of discovering, designing and manufacturing new molecules for drugs is quite complex, and this has led to exploiting AI to find such molecules, said Dr Alok who is also the author The Fourth Industrial Revolution and 100 years of AI (1950-2050). Providing an overview of where AI is being used in drug discovery and manufacturing, he said, it covers from prediction of protein structure, drug design to assessing drug–protein interactions and repurposing of drugs.
AI is also used in Polypharmacology to understand the tendency of a drug molecule to interact with living tissues and producing off-target adverse effects. Since many proteins are involved in a disease and since their eventual folds are crucial in determining as to how they would behave, it is crucial to predict their eventual folds. Fortunately, Deep Learning Networks, AlphaFold2 and RoseTTAFold can now predict the eventual folds quite accurately. The next step is to find one or more compounds that have low or no toxicity, which can potentially bind with these disease-causing proteins and neutralize them. Recently, AlphaFold2 has been used to find such compounds, and although in a nascent stage, this research looks promising, he said.
Researchers are exploiting the rules of organic chemistry and retrosynthesis, which form part of AI expert systems and coupling them Deep Learning Networks to speed up the process of drug discovery, design, and manufacturing. Repurposing an existing drug helps pharmaceutical companies in skipping phase I clinical trials, thereby, saving substantial costs. Not only Deep Learning Networks (DLNs) but also other AI algorithms are being increasingly used for understanding the association of drugs and diseases. For example, in a recent study, DLNs were used to repurpose existing drugs with proven activity against SARS-CoV, HIV, and Influenza viruses, and researchers concluded that thirteen of the screened drugs should be investigated further for potential development for fighting other viral diseases. To identify the toxicity of a compound, in 2016, DeepTox, an AI model, outperformed all methods by identifying static and dynamic features within the chemical descriptors of the molecules. It predicted the toxicity of a molecule based on predefined 2,500 toxicophoric features with high accuracy, he stated.
AI is being first used in clinical trial design to collate, harmonize, and reconcile disparate datasets. By using domain specific AI expert systems, researchers are harmonizing massive amounts of data to create a 360-degree view of each patient. After these are created, they are using them to train AI algorithms for better matching of patients with clinical trials, thereby improving patient pools for such trials, said Dr Alok.
|