Abstract

In order to making in silico prediction of Antifreeze properties in proteins,  the statistics about a protein sequence and structure such as number and type of amino acid, secondary structural contents as well as presence of k-letter words in the sequence of 14 representative proteins were extracted and analyzed, The basic codes in language for information exchange are letters and that for nucleic acids and proteins are nucleotides and amino acids, respectively. Just as speaking which is created by combination of letters and that of words in the form of sentences, we can imagine the sequence of proteins as reservoir of sentences in biology, in which the basic codes is 20 amino acids rather than letters. Based on this assumption, we investigate the presence of different k-letter words in antifreeze protein sequences using Parsprot program which was written previously by authors. Upon extraction and analysis of above mentioned parameters, they were used for artificial neural network. We then test the results for antifreeze proteins as well as other proteins and found that network detects the structural and sequence features of antifreeze proteins, with a low level of false positives (<10%). These predictions may be used in directing experimental works.