DNAbinder is a webserver developed for predicting DNA-binding proteins from their amino acid sequence using various compositional features of proteins. The SVM models have been developed on following datasets using following protein features.

Datasets: The SVM models were developed on Main dataset (having 146 DNA-binding protein chains and 250 non DNA-binding proteins), alternate dataset (1153 DNA-bindig proteins and 1153 non DNA-binding proteins) and on Realistic dataset (146 DNA-binding protein chains and 1500 non DNA-binding proteins).

Protein features: We developed two SVM models on each dataset using amino acid composition and evolutionary information.

  1. Composition based SVM model: This model has been developed using amino acid composition of proteins.
  2. PSSM based SVM model: This model has been developed using evolutionary information in form of PSSM profile obtained from PSI-BLAST search (three iterations with e-value cut-off 0.001) against nr database.
The Performance of SVM models on main/realistic datasets using amino acid composition (AAC) and PSSM profile of proteins
DatasetInputSensitivity (%)Specificity (%)Accuracy (%)MCC
Main Dataset (DNAset)AAC78.1180.8079.800.58
Alternate Dataset (DNAaset)AAC72.5172.3372.420.45
Realistic Dataset (DNArset)AAC47.9593.3389.310.40

DNAbinder allows user to submit more than one sequence for predicting DNA-biding proteins using composition based SVM model. In case of PSSM based SVM model server allows to predict one sequence at a time. If user submit more than one sequence, only first sequence will be considered for prediction.

The probability of correct prediction directly depends on the threshold. For prediction with high confidence (less probability of false positive prediction) high threshold should be choosen.

If you are using this server please site:Identification of DNA-binding proteins using support vector machines and evolutionary profiles. Manish Kumar, Michael M. Gromiha and Gajendra PS Raghava, BMC Bioinformatics, 2007, 8:463