RoseTTAFold: A software to predict protein structures using deep learning

Varuni Ankolekar, Quartesian

Lately, Artificial Intelligence has facilitated the discovery of much seamless software which has reduced the manual and cumbersome work. This has also helped in the field of Biochemistry and Structural biology. Now, the research team from the Institute for Protein Design at the University of Washington School of Medicine, Seattle has revealed a software known as ‘RoseTTAFold’ to predict protein structure rapidly with high accuracy.

With the implementation of Deep learning, the discovery of tools such as AlphaFold and trRosetta was earlier possible. These methods have surpassed traditional methods of prediction. Understanding the protein structure has been a longstanding challenge. However, is significant to determine the structure as it helps to predict its action, which could find ways to affect, modify, or control it. Determination of the structure of macromolecules is just one piece of a gigantic riddle: a crucial challenge is to connect the structural evidence to its biological function. From this standpoint, this method can influence the future of structural biology. RoseTTAFold could be a robust method as it surmounts these challenges.

Building of Network architecture:

Intrigued by available methods and to provide better approaches, they began the research with a “two-track” network approach where the information of 1D sequence alignment track of amino acids and a 2D distance between amino acid matrix tracks were utilized. However, the incorporation of the 3D coordinate level along it led to a more precise prediction method.

To predict protein structure, the data of amino acids at the 1D amino acid sequence level, the 2D distance map level, as well as the 3D coordinate level were transformed and incorporated in a three-track network. In this method, information of 1D amino acid sequence information, the 2D distance map, and the 3D coordinates flow to and fro that helps the network to together explain relationships within and between amino acid sequences, distances, and coordinates. The team used RoseTTAFold to compute hundreds of new protein structures, including many poorly understood proteins from the human genome. 

Advantages of RoseTTAFold in elucidating protein structure: 

Over the past several years, X-ray crystallography and cryo-electron microscopy (cryo-EM) have been prominent methods for elucidating protein macromolecular structures. However, RoseTTAFold can outperform the limitations of these methods. This method has an added advantage as it also facilitates the relation of structural evidence to biological function. It could quickly generate precise models of protein-protein complexes. As this approach involves the processing of information at three levels i.e., sequence, distance, along with coordinates, it facilitates overcoming difficulties that extend from cryo-EM structure determination to designing protein. Also, this method is freely accessible. 

As this method speeds up the process, it is now utilized by Scientists from all over the globe to build protein models. The Deep learning models and related scripts to run RoseTTAFold available at GitHub have been downloaded by over 140 independent research teams since July.

RoseTTAFold, which uses three-track networks could be one of the best methods to elucidate protein structure. A better understanding of protein shapes helps in the discovery of new treatments for several health disorders and could speed up the process.

Also read: The Potential of AI in Drug Discovery and Development

Reference:

1. Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., Millán, C., Park, H., Adams, C., Glassman, C. R., DeGiovanni, A., Pereira, J. H., Rodrigues, A. V., van Dijk, A. A., Ebrecht, A. C., … Baker, D. (2021). Accurate prediction of protein structures and interactions using a three-track neural network. Science, eabj8754. https://doi.org/10.1126/science.abj8754

• The Corrosion Prediction from the Corrosion Product Performance
• Nitrogen Resilience in Waterlogged Soybean plants
• Cell Senescence in Type II Diabetes: Therapeutic Potential
• Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
• AI Literacy in Early Childhood Education: Challenges and Opportunities