AI aids in drug target discovery

The long process of drug development has roughly five stages:

• Find the target, that is, the site where the drug acts and binds in the human body;
• Design, synthesis and screening of drugs;
• Pre-clinical tests to measure the effectiveness and safety of drugs;
• Clinical trials;
• Drug approval and being marketed.

The hardest part in finding a therapeutic target is understanding the pathogenesis. Before the advent of AI, it often took several generations of scientists to relay to find the answer. Taking chronic myeloid leukemia as an example, it took nearly 30 years from observing cells and discovering chromosomal mutations to understanding the treatment mechanism and finding drug targets.

Now, with the closer link between AI and medical field, we can build a library of research papers, experimental data, medical patents, and clinical medical records to quickly sort out information needed. For example, if we want to look for the difference between patients and normal people at the molecular level so as to locate the treatment target, we can integrate 200 genomics databases where various data such as published articles, trials, and clinical data are collected, and use AI to quickly give a list of possible targets. Some AI companies can also screen targets from the perspectives of immune system, signaling pathways, and molecular stereostructures.

AI shortens the design and synthesis cycle

What follows the target identification is the drug design process. In recent years, although more and more new antibodies, proteins and nucleic acids have appeared on the market, the most mainstream drugs are still small molecules synthesized in the laboratory. Because the relative mass of small molecules is small enough, it can fully bind to the target in the body. At the same time, it has advantages like good stability, long effect period, low synthesis cost, and can be taken orally.

However, the development of small molecule drugs is no easy task. The risk of randomness in the R&D process is too high, and it is hard to predict side effects and drug toxicity. In addition, if the drug itself carries too many different functional groups, it is very difficult to design and synthesize.

In response to these challenges, some AI pharmaceutical companies use three modeling tools to overcome the above difficulties in small molecule design. The first model is used to find the best drug target; the second model can design molecules that act on the target and select the most suitable one; the last model is used to find out how to synthesize these molecules in a faster pace and in a more reliable way.

Virtual test to save the time of actual pre-clinical test

After the drug is synthesized, its effectiveness and safety must be tested. Effectiveness is to evaluate the effect of the drug binding to the target; for safety, it is necessary to observe whether the drug will affect the function of other normal proteins. Usually this step is to pick out the most effective and least toxic medicine from the target candidates screened in the previous session.

Some AI tools can predict drug response. By learning millions of experimental data and thousands of protein structure information, this tool can predict the binding reaction and final effect between small molecules and target proteins. Based on the predicted results, the structure of small molecules can be further optimized to minimize toxicity. Virtual prediction can help researchers screen out less safe drug molecules, and greatly save the actual pre-clinical test. In addition to predicting the combined effect and evaluating toxicity, it is also necessary to predict drug metabolism. Some AI systems can accurately predict the metabolic processes of drugs, including: absorption, distribution, metabolism, and excretion (ADME). ADMET Prediction can help people understand the migration path of drug molecules in the body, eliminate undesirable drug candidates in advance, and shorten drug test time.

AI help optimize clinical trials

If the drug passes the preclinical cell drug test and animal drug test, it will enter the clinical stage. If a drug cannot pass a clinical phase III trial, pharmaceutical companies that have already invested huge amounts of money will seek to re-examine the experiment and find out the reason for the failure from the massive data. In order to meet the needs of pharmaceutical companies, some AI data analysis companies specifically provide clinical data analysis services for failed Phase III drugs. They will first communicate with pharmaceutical companies about their current pharmaceutical project status and requirements, and list the types of data required for analysis. After signing the confidentiality agreement, the pharmaceutical company passes the data to the AI ​​data analysis company, and after 90 days, the latter can provide an analysis report for pharmaceutical companies to optimize Phase III trials.

Drug research and development is a long and complicated process, which is full of unpredictable uncertainties. The role of AI is to eliminate as many uncertainties as possible and at the same time, help us optimize all aspects of it.