AFP-LSE

AFP-LSE is a machine-learning-based approach to predict antifreeze protein (AFP) sequences, leveraging the power of latent space learning through deep auto-encoder methods. Antifreeze proteins are critical in enabling species to survive in sub-zero environments. Despite their diverse applications across various industries, their identification remains challenging due to the low sequence and structural similarity among different types of AFPs, including those found in fish (Type I, II, III, IV, and antifreeze glycoproteins (AFGPs)).

The complexity of accurately predicting AFPs due to their subtle resemblance has limited the effectiveness of simple search algorithms like BLAST and PSI-BLAST, necessitating more reliable prediction methods. AFP-LSE addresses this need by employing a deep auto-encoder for latent space pruning followed by a deep neural network classifier. This innovative approach learns the non-linear mapping between the protein sequence descriptor and class label, significantly improving the reliability of AFP prediction.