AI and Drug Discovery 

Posted byimperialbiosciencereviewPosted inBiotechnology

By Asia Lie 

Artificial intelligence, or AI, is a technology that can interpret and learn from data it is fed to make decisions for a certain function independent of human control1. For example, some AI can create art based on a few inputs such as style or subject. It does so through many algorithms and systems, the most prominent being machine learning (ML). ML allows a computer to learn from data by picking out certain patterns and optimizations. It can be applied to form artificial neural networks, which mimic the neurons of a human brain in that they can receive inputs and eventually output something through a series of links and algorithms1. These neural networks have given rise to a new field known as deep learning2. 

AI can be used in many industries, but its applications in the pharmaceutical industry are especially interesting. Here, they can assist in a variety of things from the business aspect of drug products to the development of the products themselves. Drug discovery is a notoriously long and expensive process. It can take over a decade at a cost of $2.8 billion on average1. AI can help by recognizing molecule structures with potential and optimizing and evaluating them at a faster pace than a human. By predicting the properties and effects of certain molecules, AI could help make drug discovery faster, safer, and less costly, improving the standard of healthcare for a multitude of diseases1,3.  

The first aspect of drug discovery that AI can assist in is drug screening. By learning from massive data sets regarding molecular structure and associated properties, AI can predict the physicochemical properties of drug molecules, which can tell us about how the drug might function in a real patient1. From here, experiments can be prioritized based off of the desired properties needed2. Along with these properties, AI might be able to tell us if a drug may be toxic or bioactive, informing us of the safety of a molecule without needing to test it on live subjects. With this information, the cost and time to develop life-saving drugs could be shortened significantly. 

AI can also have an impact on designing the drug molecules themselves. Designing the optimal molecule structure is an arduous task, and most importantly involves finding the structure of the target for the drug. Often this is a protein that is overexpressed in diseases. With the data that AI has at its fingertips, it can predict the structure of target proteins and the effects that a product might have on them, giving us insight into the safety and function of a drug before it is even synthesized1,4. It can also help look for a target gene5. This would save money and time in the task of finding a functional drug to affect a protein in the desired way. One of the most important tasks AI can complete is the investigation of drug-protein interactions. Predicting drug-protein interactions allows us to know the efficacy of a drug and avoid unwanted side effects. It can also be used in drug repurposing to advance clinical trials and speed up the drug discovery process1,5.  

Beyond screening and design, AI can help with image analysis during clinical trials. Images are often key to discovering how a drug works by using microscopy to investigate the structure of a drug and the response of a protein or other molecule2. Dyes are often used to highlight certain structures in drug-protein interactions, for example in investigations with anti-cancer molecules and treatments. It is difficult to manually investigate and identify small structures and interactions in detailed microscopy images, so using an AI model can make the process much more efficient and possibly avoid human error.  

The limitations of AI in drug discovery are still numerous. One challenge is with the data itself, namely its scale, diversity, and uncertainty1. Designing a model to handle all the different factors that go into drug discovery and drug function is an obstacle itself, but the data collected has also undergone different experimental parameters and has been recorded in different formats and databases4,5. Standardizing may be necessary for use in AI but would be a long and tedious task. When there is a lack of information available, deep-learning and AI have been known to struggle with debugging and smaller data sets6. Meta-learning, a method that aims to develop models that can adapt to new tasks or problems, has been promising to solve problems when there is limited information given6. AI has also been criticized for its high computational cost, especially when it uses deep learning methods5,6. One of the biggest limitations to AI is the uncertainty in the answers it provides. Some models can return the correct answers to certain problems but without the correct logic behind them, which has become a concern6. Additionally, protein structure databases and information about drug interactions will need to be investigated further before AI models can be used to their best ability. Some methods for monitoring and explaining the decisions that AI models make will need to be developed and put into place.  

AI can have impacts in many fields, including finance, entertainment, and healthcare5. In the field of drug discovery, which is inefficient, time-consuming, and costly, AI can have many impacts, allowing us to discover drug molecules faster with the desired properties at less cost, including a decreased cost of life by reducing the need for more living subjects. It can help reduce human error and give us more insight into the properties of drug designs, hopefully allowing us to achieve treatments for complex diseases such as cancer or Alzheimer’s much sooner. 

References:

  1. Paul D, Sanap G, Shenoy S, Kalyane D, Kalia K, Tekade RK. Artificial intelligence in drug discovery and development. Drug Discovery Today [Internet]. 2020 Oct;26(1). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7577280/ 
  1. Walters WP, Barzilay R. Critical assessment of AI in drug discovery. Expert Opinion on Drug Discovery. 2021 Apr 19;16(9):1–11. 
  1. Chan HCS, Shan H, Dahoun T, Vogel H, Yuan S. Advancing Drug Discovery via Artificial Intelligence. Trends in Pharmacological Sciences. 2019 Aug;40(8):592–604. 
  1. David L, Thakkar A, Mercado R, Engkvist O. Molecular representations in AI-driven drug discovery: a review and practical guide. Journal of Cheminformatics. 2020 Sep 17;12(1). 
  1. Bijral RK, Singh I, Manhas J, Sharma V. Exploring Artificial Intelligence in Drug Discovery: A Comprehensive Review. Archives of Computational Methods in Engineering. 2021 Oct 12;29(4):2513–29. 
  1. Jiménez-Luna J, Grisoni F, Weskamp N, Schneider G. Artificial intelligence in drug discovery: recent advances and future perspectives. Expert Opinion on Drug Discovery. 2021 Apr 2;16(9):1–11.