Accurate Prediction of Fish Antifreeze Proteins Using Artificial Neural Networks

Abstract

Antifreeze proteins (AFP) prevent the growth of ice-crystal in order to enable certain organism to survive under sub-zero temperature
surroundings. These AFPs have evolved from different types of proteins and having very different sequences and structure. But they all perform the same function and become the classical example of convergent evolution. Inspired by the success of machine learning algorithms we used ANN for their prediction. A feature vector was prepared using different physicochemical property groups of amino acids, amino acid composition and dipeptide composition. Though our AFP dataset was small, the ANN is able to correctly classify the AFPs and non-AFPs. A larger dataset, incorporation of structural information and better selection of amino acid physicochemical properties for making the feature vectors will further validate better accuracy in prediction of AFPs using ANN.