We’ve all seen that generative AI tools like ChatGPT, or Stable Diffusion can create amazing text and images that very closely resemble those created by humans.
But did you know there are many other potential use cases – including incredible things like potentially helping to cure cancer?
Thanks to the speed and power of machine learning, it can be used to very quickly analyze vast datasets of chemical compounds, pharmaceutical trials and clinical outcomes. This allows researchers to speed up the difficult task of shortlisting potential candidates for testing.
Demonstrating this, Oxford-based biotech firm Etcembly recently achieved a world-first, creating an immunotherapy drug with the help of generative AI.
Let’s take a look at what this breakthrough could mean and how generative AI could soon become a key weapon in the fight against cancer.
Immunotherapy And Cancer
While some cancer treatments like chemotherapy work by attacking cancer directly, immunotherapy attempts to boost the body’s defense systems so they are more effective at destroying cancerous cells.
Etcembly’s work centers around a class of immunotherapy drugs known as T-cell engagers – designed to bring immune cells present in white blood cells closer to cancerous cells so they can do their job of killing them.
The project, known as ETC-101, does this by targeting PRAME (Preferentially Expressed Antigen in Melanoma) – a protein frequently found in cancerous cells but uncommon in healthy tissue.
The idea is that the body's natural defenses can be steered toward the areas where they will have the greatest impact while minimizing the damage caused elsewhere in the body. This means fewer side effects, as well as hopefully speedier recovery times for patients.
How Was Generative AI Used?
Etcembly created its own generative AI engine, called EMLy, based on a generative large language model (LLM) – the same technology that powers tools like ChatGPT.
EMLy was used to scan the genetic code for T-cell receptors – the molecular mechanism of the body’s immune system cells that helps them to spot foreign or abnormal entities such as a virus or cancer cell. By scanning hundreds of millions of these codes, it’s able to determine which cells are most likely to be effective at tackling specific cancer cells. In particular, it’s looking for candidate white blood cells that are most likely to have a low pM affinity, meaning they are the cells that are most likely to form bonds with the cancerous cells and destroy them.
Additionally, it also looks for cells that are less likely to be harmful to nearby healthy cells, meaning they are less likely to cause harmful side effects.
One way to imagine this is that every cancer cell has a unique lock, and the white blood cells are like a pile of keys. Find the key that fits the lock, and the cell can be destroyed – but unfortunately, there are hundreds of millions of keys that need to be tried!
Sorting through all of the possible combinations – even traditional computer simulations – would take a long time, possibly longer than the patient has. But with a generative AI like EMLy, new TCR sequences can quickly be created and tested in simulation based on all of the existing TCRs and patient outcome information held in its training data.
Where Next for Generative AI in Immunotherapy Drug Discovery?
Work combining generative AI with the search for immunotherapy treatments has the potential to spur the development of personalized medicines. These are treatments specifically tailored to individual patients, created by determining the TCR sequences most likely to be effective against a person’s specific cancer cells.
Etcembly – which received help and support from Nvidia’s Inception startup incubator – is now looking for partners that it can collaborate with in order to create more innovative treatments with its EMLy technology.
There are certainly ethical considerations that must be taken into account. Bias in training data has the potential to lead to treatments that are less effective for specific groups of people, potentially leading to disparity in healthcare outcomes. And because methods such as these rely heavily on personal data, including the genetic data of individuals, great care needs to be taken over data privacy and security.
However, by demonstrating the potential of integrating LLM technology with complex medical drug discovery, it’s hoped that it will speed up the transition of groundbreaking treatment candidates from the laboratory to the clinic and ultimately better cancer survival rates. It also has the potential to dramatically reduce costs - frequently an inhibiting factor with drug discovery.
In the near future, it’s likely we will see breakthroughs in using generative AI to create compounds for treating many other conditions. And this will only accelerate as generative algorithms become more powerful and sophisticated with time.