NPBIP: predicting binding preferences of uncharacterized nucleic-acid-binding proteins

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Nucleic-acid-binding proteins (NBPs) regulate gene expression by recognizing specific DNA or RNA sequences.
• Predicting NBP binding affinities for novel sequences is challenging due to limitations in current computational methods.
• Existing tools often require query-specific experimental data or are restricted to short sequence predictions.

Purpose of the Study:

• To develop a novel computational method for predicting the binding affinity of any query NBP to any RNA or DNA sequence.
• To overcome the limitations of existing methods that require prior experimental data or are restricted to short sequences.
• To provide a robust and accurate tool for NBP-DNA and NBP-RNA interaction prediction.

Main Methods:

• Integration of a similarity-based method and a deep-learning model (large protein language model with a hybrid convolutional-transformer network).
• Training and evaluation on 420 RNA-binding protein (RBP) and 464 DNA-binding protein (DBP) experiments.
• Utilizing publicly available source code and datasets for reproducibility.

Main Results:

• The New Protein Binding Intensity Predictor (NPBIP) significantly outperformed its individual components and competing baselines.
• Achieved mean Pearson correlations of 0.414±0.20 for RNA-binding and 0.581±0.22 for DNA-binding experiments.
• Demonstrated biological validation through recovery of canonical binding motifs for both RBPs and DBPs.

Conclusions:

• NPBIP offers a significant advancement in predicting NBP binding affinities across diverse nucleic acid sequences.
• The method provides statistically comparable or superior performance to experimental upper bounds.
• Interpretability analysis confirms the biological relevance of NPBIP's predictions.