An application software for protein secondary structure prediction based on peptide triplet units and artificial neural networks: Protein secondary structure prediction from amino acid sequences

On the basis of a bank of tendentious factors of tripeptide units, a protein secondary structure prediction system (PSSP) was built. Our research results revealed that PSSP represents a higher prediction accuracy of alpha-helix up to 89.45% on average, even the mean correct rate of alpha-helix also achieved 67.78% for all-beta proteins. PSSP only achieved a whole prediction accuracy of 59.46% for total proteins on average, higher than Chou-Fasman method. This system gave a whole accuracy of 72.64% for all-alpha folding proteins but 39.44% for all-beta proteins due to the limited data of extended conformation in train set, the absence of long-range effect, the neglect of hydrogen bridges, and losing sight of specific pairing of complementary charges and the constructive periodicity, whereas only considers conformation biases of tripeptide based on statistics analysis. However, the improved PSSP method availably advances the prediction accuracy, especially all-beta proteins up to 57.92% but all-alpha folding proteins down to 65.30%. PSSP method will play an important role in protein folding, folding codons, molecular design, and structural proteomics.