PREDICTION OF ZINC-FINGER DNA-BINDING PROTEIN

Using the neural network algorithm with back-propagation training procedure, we analysed the zinc finger DNA binding protein sequences. We incorporated the characteristic patterns around the zinc finger motifs TFIIIA type (Cys-X(2-5)-Cys-X(12-13)-His-X(2-5)-His) and the steroid hormone receptor type (Cys-X(2-5)-Cys-X(12-15)-Cys-X(2-5)-Cys-X(15-16)-Cys-X(4-5)-Cys-X(8-10)-Cys-X(2-3)-Cys) in the neural network algorithm. The patterns used in the neural network were the amino acid pattern, the electric charge and polarity pattern, the sidechain chemical propel ty and subproperty patterns, the hydrophobicity and hydrophilicity patterns and the secondary structure propensity pattern. Two consecutive patterns were also considered. Each pattern was incorporated in the single layer perceptron algorithm and the combinations of patterns were considered in the two-layer perceptron algorithm. As for the TFIIIA type zinc finger DNA binding motifs, the prediction results of the two-layer perceptron algorithm reached up to 96.9% discrimination, and the prediction results of the discriminant analysis using the combination of several characters reached up to 97.0%. As for the steroid hormone receptor type zinc finger, the prediction results of neural network algorithm and the discriminant analyses reached up to 96.0%.