Neural network applications

Neural network deals with biological problems of today. Neural networks were fist used in predicting ribosome binding sites depending on the amino acid sequence input. Stromo and his co-workers applied algorithm to sort out this prediction. They succeeded by finding translational binding site within the sequence of E. coli with the help of perceptron without any hidden unit.

The real application came into existence after the development of back propagation algorithm for perceptron which are multilayered in the year 1986. It is used to predict protein secondary structure, associated phonemes encoding, binding sites, cleavage sites and post translational modification. Network architecture plays a major role in application of protein and nucleic acids. The major applications that are discussed in detail are as follows.

• Sequence coding and output interpretation
• Sequence correlation
• Protein secondary structure prediction
• Evolutionary composition and amino acid composition prediction

Sequence coding and output interpretation : Input interpretation is of basic importance. It is possible with the help of simple linear models to solve the input representation if it is clear. The input representation transforms the information as output units with the help of the activity pattern of MLP. Coding of monomer sequence is feasible with physical and chemical features. Input is associated with window size of sequence segments. They are arranged in symmetrical manner where upstream and downstream align themselves to be of the same size. At times, asymmetrical sequence performs better than symmetrical ones.

Sequence correlation: It is obvious that proteins are highly conserved. Amino acid location helps in prediction of protein conservation either as local or global by sequence cooperativity. For instance, kinase helps in depiction of phosphorylation site motif. Correlation between each position of sequence is same as brain information retrieving process. This is possible with the help of hidden Markov model.

Protein secondary structure prediction: Classic problem of computation molecular biology is prediction of protein secondary structure involving sequence correlation, sequence coding and output interpretation. It gives information on folding of protein, motif domain and protein backbone structure by graphical representation.

Evolutionary composition and amino acid composition prediction: Chou Fasman rules helps in knowledge based networking. Network architecture performs well with Chou Fasman rules compared to Qian Sejnowski architecture. MLP architecture is evaluated with the help of Bayesian method. Neural network predicts accurate evolutionary informations by alignment methods, length dependent threshold and 3D homologues. Rost and Sander developed secondary structure prediction method called PHD architecture.

Other applications of neural network

Apart from above mentioned applications neural network also helps in

• Adaptive coding
• Position specific scoring matrix for secondary structure of protein
• Prediction of signal peptide
• Prediction of cleavage sites
• Prediction of evolutionary history of gene
• Eukaryotic gene finding
• Intron splice site prediction
• Gene structure prediction by sensor integration
• Global and local sequence information retrieval
• Prediction performance evaluation