USC Researchers Develop AI Model to Improve Protein-DNA Binding Prediction

A team at the University of Southern California has created an AI model called Deep Predictor of Binding Specificity (DeepPBS). This model, described in a recent Nature Methods paper, predicts protein-DNA binding with high accuracy across diverse protein types, potentially accelerating drug development and medical treatments. Specifically DeepPBS is a geometric deep learning tool that forecasts binding specificity from protein-DNA complex structures, allowing researchers to input data into an online computational platform.

According to senior author Remo Rohs, DeepPBS eliminates the need for high-throughput sequencing or structural biology experiments, offering a faster alternative. The AI model analyzes chemical properties and geometric contexts, producing spatial graphs that depict protein structure and its interaction with DNA. Unlike existing methods, DeepPBS can predict binding specificity across various protein families, providing a universal tool for researchers.

DeepPBS complements advancements in protein-structure prediction, such as DeepMind’s AlphaFold, by predicting specificity for proteins without experimental structures. This capability could expedite the design of new drugs, improve treatments for genetic mutations and cancers, and drive innovations in synthetic biology and RNA research.