Advances in the AI-powered simulation of proteins are making it easier than ever to discover new drugs. Neil Torbett, CEO of PhoreMost, outlines how the firm’s protein screening technology overcomes current limitations of protein structure predictions, with big applications in the emerging field of protein degradation.
Proteins are a critical component to a cell’s activities, be it sending messages, creating new molecules, or destroying old molecules. These biological machines are made up of strings of amino acid building blocks that fold themselves into complex shapes that impact their ultimate function. Predicting how these proteins fold based on their sequences has been seen as a holy grail in biology for decades as it could speed up the development of new drugs.
AlphaFold is one of the biggest recent breakthroughs in this search. This artificial intelligence (AI) system, developed by Alphabet subsidiary DeepMind, made shockwaves in late 2020 when it was able to predict the structure of many proteins accurately in minutes. When the platform became free to use last year, it captured imaginations across many areas of biology research.
“We’re getting increasingly enhanced predicted [protein] structures and the referencing to real structures is looking pretty good,” said Neil Torbett, who earlier this year became CEO of the U.K. drug discovery player PhoreMost. “I think what is really interesting for us is the translation of that information into new drug discovery programs.”
Last year, DeepMind spun off Isomorphic Labs to develop new drugs based on AI methods. However, despite the growing excitement around AlphaFold, there are limitations to the software in drug discovery.
“One of the central problems that we are thinking about is: you have structures of every single protein,” Torbett explained. “But what that information alone can’t tell you is where on a protein the correct drug site will be to elicit a therapeutic effect. AI is just not geared up to tell you that right now.”
PhoreMost seeks to fill in the gaps with its own drug discovery technology. The firm collects a library of protein fragments called PROTEINi that it throws at cell models of disease in the lab. If any protein fragments produce a desired effect on the cells, the company then identifies the target that the protein fragments bound to and uses this as a basis to develop small molecule drugs for the same target.
“We can look at other species, bacterial and fungi and everything, with the idea that we can represent the maximal protein shape diversity within our libraries,” added Torbett. “We want to reveal the most diverse number of protein folds, but we can also use information from things like AlphaFold to inspire a completely novel PROTEINi library.”
PhoreMost’s drug pipeline focuses on the treatment of cancer. The firm’s lead program — partnered with Sentinel Oncology — is expected to enter clinical trials next year for the treatment of glioma. Other programs center on destroying cancer cells by attacking their ability to repair breaks in their DNA — a hot field called synthetic lethality. One of the most exciting therapeutic areas for PhoreMost is targeted protein degradation, where a cancer treatment triggers the destruction of vital proteins in tumor cells.
“I think it’s really exciting: the idea that you can destroy a target rather than inhibit it,” said Torbett. He added that there are many protein degradation players springing up, including Amphista Therapeutics in the U.K. and Kymera Therapeutics in the U.S.
The most established class of protein degraders is proteolysis-targeting chimeras, or PROTACs for short. These small molecules are a fusion of three elements: a molecule that seeks out a cancer target; a molecule that triggers an executioner protein called an E3 ligase to destroy the target; and a linker that connects the two. Most PROTACs in development only have a choice of two E3 ligases to recruit; if tumor cells become resistant to these ligases, then many PROTACs wouldn’t work well.
“The field has really struggled to open up the ligase space,” noted Torbett. “There are over 500 ligases in the cell. There is a gold rush to uncover new ways to degrade therapeutic targets.”
PhoreMost aims to accelerate research into other ligases to make it harder for cancers to resist protein degradation drugs. The firm first identifies protein fragments that can bind to ligases and degrade cancer targets, and then it searches for small molecule candidates that can have the same effect.
“We’ve been able to unpick the mini protein-ligase interfaces such that we are now in drug discovery on six different E3 ligases,” added Torbett. “We can find the small molecules that would be candidates for binding the site that we found. So that’s a really exciting way of accelerating small molecule hit-finding and discovery.”
PhoreMost’s lead ligase in preclinical development is aimed to be effective in cancers where resistance to current PROTACs may emerge. In addition to ongoing partnerships with Boehringer Ingelheim and Oxford Biomedica, PhoreMost anticipates signing big pharma drug development deals in the coming months.
AI approaches are very useful for guiding PhoreMost’s drug discovery techniques, yet Torbett believes wet lab techniques remain crucial to unearthing new ways to bind to disease targets. Without wet lab approaches, Torbett predicts that AI players will be “somewhat limited to circling around the known druggable space.”
“They’ll be able to prioritize or look at active sites on well drugged targets,” he added. “They can go make more kinase inhibitors. But what they can’t systematically do is really expand the druggable universe.”
Nonetheless, big advances are expected in the coming years with the use of AI in protein mapping. In particular, the accuracy and resolution of protein structure is improving incrementally, and one day, fully virtual screens could be carried out of binding pockets on target proteins.
“That is really transformational,” concluded Torbett. “For me, that’s the massive step.”