A new artificial intelligence model called Deep Predictor of Binding Specificity (DeepPBS) has been developed by USC researchers and published in Nature Methods. This model can predict how different proteins bind to DNA accurately, across various types of proteins. This advancement in technology has the potential to reduce the time required to develop new drugs and medical treatments significantly. DeepPBS is a geometric deep learning model designed to predict protein-DNA binding specificity based on the structure of protein-DNA complexes. By inputting this data into an online computational tool, scientists and researchers can analyze and understand the binding specificity of proteins to different DNA sequences.
Proteins in DNA complexes are typically bound to a single DNA sequence, but for better understanding of gene regulation, it is crucial to have access to the binding specificity of a protein to any DNA sequence or region of the genome. DeepPBS eliminates the need for high-throughput sequencing or structural biology experiments to determine protein-DNA binding specificity. The AI tool captures the chemical properties and geometric contexts of protein-DNA interactions to predict binding specificity accurately. By producing spatial graphs that illustrate protein structure and the relationship between protein and DNA representations, DeepPBS can predict binding specificity across various protein families, which sets it apart from existing methods limited to specific protein families.
The development of DeepPBS marks a major advance in protein-structure prediction. With the rapid progress in this field since the emergence of DeepMind’s AlphaFold, which predicts protein structure from sequence, scientists and researchers now have access to a vast amount of structural data for analysis. DeepPBS can be used in conjunction with structure prediction methods to predict specificity for proteins lacking experimental structures. This new tool has various applications, including accelerating the design of drugs and treatments for specific mutations in cancer cells, facilitating new discoveries in synthetic biology, and enabling advancements in RNA research.
The study on DeepPBS was conducted by a team of researchers from USC, University of California, San Francisco, and University of Washington. The team included authors such as Remo Rohs, Raktim Mitra, Jinsen Li, Jared Sagendorf, Yibei Jiang, Ari Cohen, Tsu-Pei Chiu, and Cameron Glasscock. The research was primarily supported by NIH grant R35GM130376. The applications of DeepPBS in drug development, cancer research, synthetic biology, and RNA research demonstrate the potential impact of this AI tool in advancing medical treatments and scientific discoveries.
Overall, DeepPBS represents a significant milestone in the field of protein-DNA binding specificity prediction. By leveraging geometric deep learning methods, this AI tool provides accurate predictions across different types of proteins, offering a universal approach for researchers. With the potential to revolutionize drug development, cancer research, and other areas of study, DeepPBS promises to streamline processes and accelerate the pace of discovery in the biomedical field.