Revolutionizing Protein Design: AI-Powered Nanopores Unlock New Frontiers in Molecular Biology
A research led by the VIB-VUB Center for Structural Biology (Belgium) and the University of Washington School of Medicine (USA)
The 3D structures and functions of proteins are encoded in their sequence of amino acids. Advances in molecular modelling and AI can now be leveraged to design proteins from scratch and with new-to-nature properties (de novo). However, so far de novo design research has mainly focused on protein folding in water, while 30% of natural proteins do not fold in water but in lipid membranes.
Unlocking the power of nanopores: key findings:
New Protein Design Methods – from the computer model to the molecular biology bench: Two research teams from the University of Washington and the VIB-VUB Center for Structural Biology in Belgium have created new computational methods aimed at tackling two grand challenges of protein design: the design of transmembrane nanopore proteins and that of proteins specifically binding small-molecules. They applied the methods to design new protein structures unlike any found in nature and showed that they fold and function exactly as predicted by the models.
Two complementing studies advancing nanopore applications for sensing and sequencing. The Belgian team designed synthetic beta-barrel proteins that can embed into lipid membranes and form nanoscopic holes that permeates ions, mimicking the properties of natural nanopores. The US team created a workflow to design binder proteins that attach to specific small molecules with high affinity. The two studies were nicely combined to generate a new protein nanomachine designed entirely on the computer and able to sense cholic acid (a biomarker of metabolic disorders).
Advantages over Traditional Methods. Current nanopore sensor development primarily relies on naturally occurring proteins, which are not ideal starting points because these proteins evolved to fulfill completely different tasks in the cell. De novo design can theoretically produce an unlimited number of diverse pores by precisely controlling their shape and properties on the atomic scale.
Creating a new computational model. One of the significant challenges in designing beta-barrel nanopores is the lack of computational models for predicting protein folding in lipid membranes. The team created a model based on existing knowledge of membrane proteins.
A deep learning pipeline. The US team developed a deep learning-based pipeline to design small molecule-protein binders. This method reduces the need for extensive wet lab screening, making the process more efficient.
Potential Applications and Future Directions. The teams aim to enhance the computational pipeline by incorporating fine-tuned AI methods, and to design nanopores capable of detecting multiple targets simultaneously.
What is probably important to mention is that many of the design calculations were done on the VSC Tier-1 supercomputer (grant number: lt1_2021-32 an lt1_2022-32)
Read the full publication in the American Association for the Advancement of Science (AAAS), here
Comments