AI in Drug Discovery
Developing new or more effective drugs for treating medical conditions can revolutionise care, and drug discovery is a huge part of the business of pharmaceutical companies. However, finding which drugs are effective for treating which conditions is difficult. Identifying and screening candidate drugs is typically extremely time-consuming, which makes the search for new drugs slow, uncertain, and very expensive.
In modern science, this is not for lack of data. Plenty of data exists on how small molecules interact with biological systems such as proteins. However, sorting through all this data to find promising combinations of molecules and biological pathways to treat particular conditions is very slow. Machine learning offers a way to overcome this problem.
We reported recently on Alphafold – a machine-learning tool capable of predicting protein structures with much greater reliability than previous tools. Other programs already exist that can predict the structures of small molecules, which are much easier to determine from their chemical composition than the structures of proteins. Based on the predicted structures of proteins and small molecules, machine-learning can predict their interactions, and work through libraries of molecules to identify candidate drugs much more quickly than would be possible with human effort alone.
This type of processing can identify entirely novel drugs, but may also be used to identify new applications of existing drugs. Identifying new uses of existing drugs can be particularly valuable, since manufacturing capacity and detailed data on side effects may already exist that can allow the drug to more rapidly be repurposed to treat a new condition.
Machine learning can not only identify molecules likely to interact with a target protein, but may also be able to extrapolate properties such as toxicity and bio-absorption using data from other similar molecules. In this way, machine-learning algorithms could also effectively carry out some of the early stages of drug screening in silico, thereby reducing the need for expensive and time-consuming laboratory testing.
Other applications of machine learning in drug discovery include personalised medicine. A major problem with some drugs is the varying response of different individuals to the drug, both in terms of efficacy and side-effects. Some patients with chronic conditions such as high blood pressure may spend months or years cycling through alternative drugs to find one which is effective and has acceptable side effects. This can represent an enormous waste of physician time and create significant inconvenience for the patient. Using data on the responses of thousands of other patients to different drugs, machine learning can be used to predict the efficacy of those drugs for specific individuals based on genetic profiling or other biological markers.
Identifying candidate drugs as discussed above relies on knowing which biological target it is desirable to affect, so that molecules can be tested for their interaction with relevant proteins. However, at an even higher level, machine learning techniques may allow the identification of entirely novel mechanisms for treating medical conditions.
Many studies exist in which participants have their genetic data sequenced, and correlated with data on a wide variety of different phenotypes. These studies are often used to try to identify genetic factors that affect an individual’s chance of developing disease. However, machine learning techniques can also identify correlations between medical conditions and other measurable parameters, such as expression of certain proteins or levels of particular hormones. If plausible biological pathways can be determined using these correlations, this could even lead to the identification of entirely new mechanisms by which certain conditions could be treated.
Examples of AI-based drug discovery already exist in the real world, with molecules identified using AI methods having entered clinical trials. Numerous companies are using AI technology to identify potential new drugs and predict their efficacy for individual patients. Some estimates suggest that over 2 billion USD in investment funding was raised by companies in this technology area in the first half of 2021 alone. As with any technology, patents held by these companies allow them to protect their intellectual property and provide security for them and their commercial partners.
Machine learning excels at identifying patterns and correlations in huge data sets. Exploiting this ability for drug discovery has the potential to dramatically improve healthcare outcomes for patients, and streamline the unwieldy and expensive process of developing new treatments. We may stand on the threshold of a new era of personalised medicine and rapid drug development.