The doctors had nearly given up on their patient. The 82-year-old “Paul” from Vienna was suffering from an aggressive form of blood cancer and had undergone six different chemotherapy treatments without success. With each treatment, another drug was removed from the list of potentially effective therapies, diminishing hope for a cure. The harsh reality was that standard cancer medications were ineffective against Paul’s blood cancer.
With nothing left to lose, Paul’s doctors enrolled him in a study at the Medical University of Vienna. Researchers aimed to test a new technology developed by the British company Exscientia, designed to find the best drug for each patient while considering individual biological characteristics.
The researchers took a tissue sample from Paul containing both cancerous and normal cells and divided it into over 100 smaller samples. These samples were exposed to various drug cocktails. Using robots and artificial intelligence-based imaging, they could automatically detect and document even the smallest changes in the cells.
Unlike Paul’s previous treatments, where doctors tried different drugs over months, this method allowed them to test dozens of drugs simultaneously outside the body. They could also test therapies typically deemed unsuitable for this type of cancer.
This strategy proved successful. Although the most effective drug from the tests could not be used because Paul was too weak, the second-best drug worked. This was a cancer medication from Johnson & Johnson that had not been tried in the initial treatments because previous studies showed it was ineffective for this type of blood cancer. However, the Vienna case study proved otherwise, as Paul’s cancer disappeared two years after the treatment.
For Andrew Hopkins, CEO of Exscientia, this success indicates that the new method could significantly change cancer therapy. The company plans to not only help select the right medication for individual patients but also to overhaul the entire drug development pipeline. The first two drugs developed with AI assistance are in clinical trials, and two more will soon be submitted.
Exscientia is not alone in focusing on AI. Many startups are exploring the use of machine learning in the pharmaceutical industry. AI helps predict how potential drugs will behave in the body, allowing for the elimination of likely ineffective substances early on. This reduces the need for labor-intensive lab work and speeds up drug development, which traditionally takes over ten years and costs billions.
New laboratories are being established worldwide. Exscientia opened a research center in Vienna, and Insilico Medicine, headquartered in Hong Kong, opened a large lab in Abu Dhabi. Around two dozen AI-developed drugs are currently in clinical trials or nearing that stage.
However, drug research with AI is still in its infancy. Many companies make claims they cannot substantiate. A new generation of AI companies is focusing on three key weaknesses in drug development: selecting the right target molecule in the body, developing the correct molecule to interact with it, and identifying the patients most likely to benefit.
AI is not a cure-all. Experiments in labs and human testing, the slowest and most expensive steps in development, cannot be bypassed. While AI saves time and replaces many manual steps, final validation must occur in the lab.
Machine learning is replacing manual work in drug development. Computer-assisted techniques have been used for decades, but models still require manual creation, a slow and error-prone process. Machine learning can automate the creation of complex models using vast datasets, making it easier and quicker to predict how drugs will behave in the body. Early experiments can be simulated in computers, and machine learning models can search through large pools of potential drug molecules.
Many companies use machine learning initially to identify targets in the body before simulating drug behavior. Some, like Exscientia, use language processing models to analyze data from scientific publications, extracting information and coding it into knowledge graphs. These models predict which targets are most suitable for treating specific diseases.
Selecting a target is just the beginning. The greater challenge is designing a drug molecule to interact with the target as desired, an area currently seeing the most innovation. The interaction between molecules in the body is highly complex, and drugs must navigate hostile environments like the gut to be effective.
Generate Biomedicines aims to achieve this with generative AI, similar to the technology behind text-to-image software like DALL-E 2. Instead of manipulating pixels, their software works with random amino acid strands to create protein structures with specific properties.
Absci is also developing protein-based drugs, focusing on antibodies that naturally eliminate bacteria and viruses. They redesign existing antibodies to bind better to targets using mathematical models trained on lab data. After simulation adjustments, the best-performing structures are synthesized and tested.
Apriori Bio is focusing on Covid, developing vaccines to protect against a wide range of virus variants. They create millions of variants in the lab and use machine learning to predict how antibodies will perform against potential variants.
AI opens a vast, untapped pool of biological and chemical structures for future drug development. The number of potential drug molecules is far greater than the current pool of about ten million molecules used by major pharmaceutical companies.
Verseon uses both old and new computer techniques to explore possibilities, generating millions of molecules and testing their properties. Their first drug is in clinical trials, with more to follow, targeting diseases like cardiovascular conditions, infections, and cancer.