Prediction of protein structural classes for low-similarity sequences using reduced PSSM and position-based secondary structural features

• We simplified PSSM structure with reduced amino acids.
• We analyzed long-range structural properties of the reduced PSSM.
• We explored protein position-based structural features.

Many efficient methods have been proposed to advance protein structural class prediction, but there are still some challenges where additional insight or technology is needed for low-similarity sequences. In this work, we schemed out a new prediction method for low-similarity datasets using reduced PSSM and position-based secondary structural features. We evaluated the proposed method with four experiments and compared it with the available competing prediction methods. The results indicate that the proposed method achieved the best performance among the evaluated methods, with overall accuracy 3–5% higher than the existing best-performing method. This paper also found that the reduced alphabets with size 13 simplify PSSM structures efficiently while reserving its maximal information. This understanding can be used to design more powerful prediction methods for protein structural class.