AI is becoming really widespread in the field of drug discovery. Drug discovery and development is a process that takes a long amount of time and money, and AI can help cut these costs.
The first phase of the drug development process is the drug discovery process. New drugs can be discovered by learning new aspects about a disease, new technologies, treatments and more. There can be many different possible compounds at this stage that could be possible drugs. After some studies only a few compounds are chosen to be studied further. In the pre-clinical phase, researchers study how the compound is absorbed, distributed, metabolized and excreted. They also study possible side effects, benefits, mechanisms of action, the best dosage, the best way to administer the drug, the effects the drug has on different groups, how the compound interacts with other drugs and how the compound compares to other drugs. They ultimately try to find out if the drug is too harmful to be used for further testing on humans. If the drug is proven safe enough, then it moves on to the next phase.
The next phase is the clinical research phase. This phase lasts several years and involves the testing of the drug on both healthy patients and patients with the diseases. This phase is to determine the side effects of the drug and its effectiveness before applying for approval by the FDA.
AI is used in this process in many ways. ML is an especially useful tool due to the increase in data over the years. In ML there is supervised and unsupervised learning. Supervised learning is labeled data and is usually used to predict outcomes in a clinical trial. Some examples of this involve regression and classification. Unsupervised learning has unlabeled data, and is usually used to characterize data by separating them into groups or making associations. An example of this would associating that people who have disease A, also have disease B. Natural language processing is also commonly used to read through electronic health records and other places with written data.
In the drug discovery process, AI can be helpful with predicting chemical structures of compounds that will bind to certain targets. A target is a biological entity that chemical structures bind to. AI can also identify possible targets for a compound by looking at existing effective drugs. AI can also predict how a drug will act in the human body. It can predict if a drug will have high toxicity, or will have undesirable absorption, distribution, metabolism or excretion. This is helpful because finding out earlier if a drug will fail can save money and time.
AI can also be helpful with drug repositioning. Repositioning is when a new application for a drug is found. This is good for situations where there aren't many patients for a certain disease, so a clinical trial isn't feasible or too expensive. AI compares existing drugs to new compounds to see if there are any matches, and AI also compares different diseases to find commonalities. Sometimes both these methods are combined to see if a drug can be repositioned.
AI can help with patient population recruitment as well. It can search through electronic health records to find eligible patients for a clinical trial. AI can also monitor patients, either by looking at electronic health records, or by taking data in real time through wearable devices like smartwatches.
AI can also be used for personalized treatment using statistical models. AI in this field is not as developed because a large amount of data from clinical trials is needed, and this usually isn't available. AI also looks at how gene variants effect the pharmacokinetics and make predictions about enzyme activity. Pharmacokinetics is how the organism effects a drug.
Some limitations of this is that AI can be heavily bias by the data that it is trained by. Any gender, racial, age, class etc. bias in data will show up in the AI model as well. Additionally, the more accurate a model is the harder it is to understand what the model is doing. This is an issue in clinical trials because there needs to be transparency with the patients.
For this week, I ran Kmeans clustering on data from a mental health clinical trial.
Kaggle dataset: https://www.kaggle.com/datasets/shashwatwork/clinical-dataset-of-the-cypguides-trial
Clinical Trial Clustering Code