Predicting DNA-binding Sites of Proteins from Amino Acid Sequence
Author Information
Author(s): Yan Changhui, Terribilini Michael, Wu Feihong, Jernigan Robert L, Dobbs Drena, Honavar Vasant
Primary Institution: Utah State University
Hypothesis
Can a machine learning approach effectively identify amino acid residues involved in protein-DNA interactions?
Conclusion
The study demonstrates that Naïve Bayes classifiers can efficiently identify potential DNA-binding sites in proteins using sequence information.
Supporting Evidence
- The classifier achieved 71% overall accuracy and improved to 78% with additional input features.
- In 33% of the proteins, the classifier identified at least half of the interface residues correctly.
- The classifier's predictions significantly overlap with known DNA-binding motifs.
Takeaway
Scientists used a computer program to find parts of proteins that stick to DNA, helping us understand how genes are controlled.
Methodology
A Naïve Bayes classifier was trained using amino acid identities and sequence entropy to predict DNA-binding residues.
Potential Biases
The classifier may misclassify non-DNA-binding proteins as DNA-binding due to the absence of evidence in Gene Ontology annotations.
Limitations
The classifier's performance may not improve with additional structural features, and it was trained only on known DNA-binding proteins.
Statistical Information
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website