Pilani, 27 October 2026: An international research study co-authored by Dr. Sushmita Basu, Assistant Professor at Birla Institute of Technology and Science (BITS) Pilani, has developed DNAreader, an advanced AI-driven tool designed to identify the specific parts of proteins that interact with DNA in both structured and flexible regions. The project was led by scientists at Virginia Commonwealth University (USA) and Xinyang Normal University (China) in collaboration with Hebei University of Technology (China).
Published in the prestigious journal Nucleic Acids Research, the study addresses an important challenge in understanding how proteins interact with DNA. These interactions are vital for processes such as controlling gene activity, copying and repairing DNA, and maintaining the organization of genetic material. However, identifying the exact parts of proteins involved in these interactions can be challenging, particularly in proteins that lack a fixed structure
To solve this, the research team, comprising Dr. Jian Zhang (Xinyang Normal University), Dr. Sushmita Basu (Virginia Commonwealth University, BITS Pilani), Dr. Jingjing Qian (Hebei University of Technology) and Dr. Lukasz Kurgan (Virginia Commonwealth University), developed a first-of-its-kind tool optimized to predict DNA–binding residues in both structured and disordered protein regions. Another challenge with existing DNA prediction tools is that they often incorrectly identify residues that interact with non-DNA molecules as DNA–binding residues.
DNAreader addresses this challenge through two combined innovations:
· a specially designed batch-training strategy that exposes the model to balanced examples of DNA-binding, other-ligand-binding and non-binding residues in both structured and disordered regions,
· a contrastive learning approach that sharpens the model’s ability to distinguish between these residue types.
Together, these innovations improve prediction accuracy and substantially reduce cross-predictions.
Speaking on the technology, Dr. Sushmita Basu, Assistant Professor, Department of Biological Sciences, BITS Pilani, said, “Accurately identifying DNA–binding residues is particularly challenging when proteins contain intrinsically disordered regions. DNAreader addresses this gap by bringing structured and disordered regions within a unified prediction framework. The integration of transformer-based learning with contrastive learning enables the model to distinguish DNA–binding residues more effectively while reducing incorrect predictions. We are confident this resource will support researchers in studying protein–DNA interactions and advance AI-assisted discovery in biology.”
In evaluation against 14 existing prediction tools using independent, low-sequence-similarity test datasets, DNAreader statistically outperformed all benchmarked tools, achieving an accuracy score (AUROC) of 0.86 to 0.89, compared to 0.75 to 0.79 for existing top-performing tools. The researchers illustrated these gains with real protein examples, including the human estrogen-related receptor ERRβ, where DNAreader accurately traced DNA–binding residues across structured and disordered segments, and an unrelated control enzyme where DNAreader correctly avoided false flags.
DNAreader is freely available as a user-friendly web server requiring only a FASTA-formatted sequence as input at http://biomine.cs.vcu.edu/
This high-impact publication reflects the world-class academic expertise housed within BITS Pilani as the institute continues to advance interdisciplinary research bridging artificial intelligence, computational biology, and the life sciences.
