Many drug and antibody discovery pathways focus on cell membrane proteins, which are difficult to study because they are embedded in the lipid-containing outer layer of cells and insoluble in water-based solutions. To address this challenge, a research team led by Casper Goverde at the Laboratory of Protein Design and Immunoengineering used deep learning to design synthetic soluble versions of these proteins. Traditional screening methods rely on observing cellular reactions to drug candidates or extracting membrane proteins from mammalian cells, but this computational approach allows for the production of soluble protein analogues using bacteria, which can then bind directly with molecular candidates of interest. This process is estimated to be around 10 times less expensive than using mammalian cells for production.
The researchers’ work on flipping the script on protein design involved using artificial intelligence networks to design soluble versions of key cell membrane proteins based on their existing 3D structures. They utilized the AlphaFold2 structure prediction platform from Google DeepMind to produce amino acid sequences for these proteins and then optimized these sequences for functionality and solubility using the ProteinMPNN network. This approach showed remarkable success in producing soluble proteins that maintained parts of their native functionality, even for highly complex folds that had eluded other design methods. A notable success was the redesign of a soluble analogue of the G-protein coupled receptor (GPCR), a major pharmaceutical target that represents a significant portion of human cell membrane proteins.
The success of designing a soluble GPCR shape opens up opportunities for faster and easier drug screening processes. Additionally, the team sees potential applications for their pipeline in vaccine research and cancer therapeutics. For example, they successfully designed a soluble analogue of a claudin protein, which is involved in making tumors resistant to the immune system and chemotherapy. This proof-of-concept demonstrates the pipeline’s potential to generate interesting targets for pharmaceutical development in various areas of research. The researchers’ work was recently published in the journal Nature, showcasing the innovative approaches to protein design using deep learning and computational methods.
The development of soluble, stable analogues of cell membrane proteins is a significant advancement in the field of drug and antibody discovery. By making these proteins more accessible and easier to work with, researchers can accelerate the process of screening for novel drugs and therapeutic targets. The cost-effective production of soluble protein analogues using bacteria instead of mammalian cells also has the potential to revolutionize the pharmaceutical research industry. Overall, the success of this research highlights the power of deep learning and computational methods in protein design and opens up new possibilities for drug discovery and development in various areas of biomedicine.