Over the past few years, Artificial Intelligence (AI) has taken an important role in the healthcare industry. Startups are leveraging this technology to improve healthcare services, from automating administrative tasks to diagnosing and treating diseases. One of the most promising use cases addresses the challenge of diseases caused by non-druggable proteins. These proteins have long puzzled the scientific community because they remain un-addressable despite being linked to many serious illnesses. If solved, it would represent a breakthrough for the biotech and pharmaceutical industries.

In this blog post, we'll explore various types of proteins, how errors in their formation affect our health, and how Sibylla Biotech, one of our portfolio companies, is using machine learning to help develop new treatments targeting undruggable targets.

Proteins: Function, Folding, and Disease

Although we frequently associate proteins with nutrients found in food, proteins are much more than that. They’re actually small molecules found inside cells that perform many different vital functions for our bodies. A typical human cell houses between 20,000 and over 100,000 different types of proteins. Some offer structural support to cells and tissues, others transport molecules across cell membranes, while others act as enzymes to catalyze biochemical reactions. All of this is possible because proteins have a unique common trait: folding.

Proteins start as chains of amino acids, and depending on their number and sequence, the resulting protein will fold into a specific shape. And it’s this specific conformation that will determine the protein’s function. There are 22 types of amino acids, and their sequence determines how the chain will fold. Over 200 million proteins exist, but we only know the exact 3D conformation of a fraction of these. For decades, scientists have been working on how to figure out a protein's shape just from its strings of amino acids; this is what we call the “protein folding problem.” But sometimes proteins don't fold correctly, which can lead to diseases like Alzheimer's or cancer. Genetic mutations, protein production errors, or unfavorable cellular conditions like temperature changes or other stress conditions are the main reasons for misfolding. Non-druggable proteins are highly stable entities that lack binding sites that traditional drugs can target, making thus a treatment difficult, if not impossible. For this reason, understanding the mechanisms behind proper folding is crucial, as it can help develop new strategies to prevent or manage these diseases by addressing the root causes of folding. And it’s this specific problem that Sibylla Biotech is trying to address.

Sibylla Biotech's Solution

Sibylla Biotech is a young company based in Milano, Italy that focuses on predicting protein folding through its PPI-FIT platform (Pharmacological Protein Inactivation by Folding Intermediate Targeting). It achieves this by repurposing sophisticated mathematical models used in particle physics (commonly used to analyze data from particle colliders like the one in CERN). PPI-FIT helps by identifying “druggable pockets” as proteins fold. In these pockets, certain molecules can bind and stop the protein synthesis, marking it ready for destruction and hence preventing diseases from developing. This process is known as protein “degradation,” which plays a crucial role in developing new therapies.

The primary challenge to solving the protein folding problem has been the large number of combinations and permutations possible from amino acids, typically larger than the number of atoms in the universe (known as Levinthal’s paradox). For instance, mapping out the folding of α-1 antitrypsin 1 on the ANTON-2 supercomputer (the benchmark for structural biology) using traditional molecular dynamics would have taken 400,000 years, while Sibylla's AI-based system accomplishes this in just 1.5 months!

An analogy can be drawn to Sibylla's groundbreaking technology, which is like curing a disease that affects an egg before its shell forms. This approach is transformative because there is no need to break the shell. Sibylla's model provides an innovative method to identify pockets and respective molecules that can interact with it, offering a completely new mechanism of action. This approach can tackle diseases that were previously deemed untreatable. Hence, the company focuses on targeting serious, non- druggable, protein-related diseases.

Conclusion

Sibylla is one example of how AI can speed up research and reduce costs in the pharmaceutical industry. Startups like Acodis use AI to predict how new treatments might perform by analyzing vast amounts of patients' historical data and clinical trials.

On another note, Vara.ai utilises computer vision to detect breast cancer from medical imaging accurately. This speeds up the testing/diagnostics processes and helps scientists find new treatment options more quickly.

AI is playing a crucial role in accelerating biotech, and many startups are making significant progress in improving our understanding of various diseases. Sibylla, in particular, is taking an innovative approach that could lead to a future where complex health issues can be resolved and cured more efficiently, even before they form.

About Vi Partners:

Vi Partners is the longest-established Swiss Venture Capital firm. For more than 20 years, Vi Partners has been supporting innovative Technology and Healthcare companies, investing over CHF 350m in 66 startups. Vi Partners-managed funds are backed by Switzerland's most visible companies and institutions, including ETH Zurich, ABB, Buhler, Credit Suisse, Hilti, McKinsey, Nestlé, Schindler, Sulzer, Suva, and ZKB, as well as the European Investment Fund and many other institutional investors

1 Alpha-1 antitrypsin deficiency is a rare genetic disorder that is passed on in families and primarily affects the lungs and liver.