The Future of Protein Design: A Look at Four Key Methods

Have you ever heard of the protein folding problem? It's a long-standing challenge in biochemistry, physics, and computer science that has puzzled scientists for decades. Essentially, the issue is that proteins are made up of long chains of amino acids that must fold into complex shapes, and predicting how a given protein will fold is extremely difficult. However, new innovations in machine learning are making it possible to crack this problem and design proteins with specific functions.

Let's say you're a scientist looking to design a new type of protein that can break down a specific pollutant in the environment. In order to do this, you would need to know the atomic-level details of how the protein interacts with the pollutant. Traditionally, this would involve a lot of trial and error in the lab, and could take years or even decades. But with machine learning, you can train software to predict how a protein will interact with a given target, and use that information to design a highly specific protein in a matter of weeks.

Real-life Examples

One example of this approach in action is the work being done by the protein design company, HUMA Therapeutics. They are using machine learning to design antiviral drugs that can bind to the spike protein of the COVID-19 virus, preventing it from infecting human cells. The protein design process involved screening millions of protein variants using molecular dynamics simulations and neural networks to predict which variants would be most effective.

Another company in this space is Insilico Medicine, which is using machine learning to design new drugs for a range of diseases. They have developed a platform called GENTRL that uses deep learning algorithms to generate new drug candidates, taking into account factors such as activity, toxicity, and specificity. They are currently conducting clinical trials on their first AI-generated drug, which is a treatment for idiopathic pulmonary fibrosis.

Meanwhile, researchers at Carnegie Mellon University are using machine learning to tackle the protein folding problem head-on. They have developed a software tool called ALEPH that uses neural networks to predict how a given protein sequence will fold, and have achieved accuracy levels of up to 92%. This is a significant improvement over previous methods, which were typically accurate only to around 50%.

Main Companies

• HUMA Therapeutics
• Insilico Medicine
• Carnegie Mellon University

Summary and Critical Comments

• Machine learning is revolutionizing protein design, enabling scientists to create highly specific proteins for a range of applications.
• Companies like HUMA Therapeutics and Insilico Medicine are using these methods to develop new drug candidates for a range of diseases.
• While these innovations are exciting and promising, there are still significant challenges to be overcome in terms of accuracy and accessibility. There is also concern around the ethical implications of AI-generated proteins and drugs.