AI Guided Protein Structure Prediction: What does it mean for drug discovery?
Applying artificial intelligence techniques to better understand proteins could significantly accelerate drug development, argues Nick Massouh, Biotechnology Sector Lead at Tarleton.
AI could help us better understand protein structures
Proteins are the molecular machines of the human body. They reside within all organs, tissues and cells, governing the essential biological processes that enable life. Such is the importance of their activity within the body, malfunctioning proteins are responsible for thousands of different diseases, making them crucial targets for drug discovery. Likewise, an increasing number of therapies harness the functions of proteins, such as antibodies and hormones, to suppress the effects of various diseases.
Common amongst all proteins is the intrinsic link between their three-dimensional structure and their biological function. The shape, size and conformation of proteins are adapted to enable them to carry out a range of important processes that often become perturbed in disease, ranging from the transport of substances into cells to the catalysis of vital chemical reactions. Moreover, in order to develop potent drugs against a target protein, an in-depth understanding of their structure is critical.
Unfortunately, however, solving the structure of proteins using traditional experimental methods can often be both incredibly challenging and time consuming. Thus, as is the case with many key challenges in modern times, science has increasingly turned to computation to help solve these issues.
The 2024 Nobel Prize in Chemistry was awarded to Demis Hassabis and John Jumper from Google DeepMind for the creation of AlphaFold2, and to David Baker from the University of Washington for his pioneering work in computational protein design. These innovations have revolutionized the study of protein structures using artificial intelligence (AI).
Before AlphaFold, the large majority of known proteins had no experimental structures resolved for them. AlphaFold is a neural network-based model driven by a deep learning algorithm that has been trained on vast amounts of physical and biological data on protein structures. Largely relying on homology with experimentally derived structures, energy minimisation functions and geometric constraints, AlphaFold enables protein structure prediction to near-experimental accuracy, with amino acid sequence being the only input required. Such is the impact of this technology, AlphaFold has been used by millions of scientists globally to predict the structure of more than 200 million proteins. Despite the recency of its development, it has already proven to be an incredibly valuable tool for studying disease pathology and therapeutic discovery.
The Baker lab developed the Rosetta software suite, a large collection of computer programmes that enable ab initio protein structure prediction—essentially creating entirely new proteins from scratch. Rosetta incorporates powerful AI tools into many of its methods; namely RFDiffusion, which combines structure prediction networks with generative diffusion models to produce thousands of biologically stable protein structures at once, ranging from whole protein complexes to small peptide binders. This technology could be set to revolutionise the biotechnology industry, with many of its features being applicable to drug discovery, synthetic biology and seemingly endless other opportunities.
This is the first time the Nobel Prize has recognized an AI-driven scientific breakthrough, and it almost certainly will not be the last. The impact of these technologies is consistent with the exponential increase in AI-discovered drugs within the pharmaceutical industry over the last half decade.
Drug development has traditionally been inefficient and expensive, with pharmaceutical companies and biotechs generally taking more than 10 years to bring a drug to market and spending up to billions of dollars throughout the process. AI-based discovery approaches, akin to those established by last year’s Nobel laureates, can solve these issues and optimise the drug development pipeline.
But whilst AI has the potential to transform discovery, navigating the intellectual property laws surrounding AI can be problematic if not done correctly. This is largely due to the fact that drugs developed without sufficient human contribution may not be eligible for patent protection. Whilst this adds an extra layer of complexity to the already challenging process of drug discovery, it is crucial that biotech companies working within this space communicate to investors the ways in which AI can be harnessed in research without foregoing the patentability of their drugs.
For example, biotechs developing their own AI systems, such as large language models designed specifically for their research purposes, need to convince investors that their technology is distinct from general purpose AI platforms and is therefore a strongly eligible patent opportunity. Likewise, biotech companies that utilise AI as a mechanism of drug design that goes hand-in-hand with experimentally-guided material modification must ensure investors understand their platform fulfils the human contribution requirement of patenting.
Undoubtedly, AI advancements such as AlphaFold and Rosetta will help pave the way for a future where drug discovery is more precise, efficient and impactful. However, as the biotech industry increasingly leans towards AI-guided therapeutic development, a catered and stringent communications strategy is essential for securing long-term investment within this space.
