DBP-iDWT: Improving DNA-Binding Proteins Prediction Using Multi-Perspective Evolutionary Profile and Discrete Wavelet...

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• DNA-binding proteins (DBPs) are essential for DNA replication, recombination, and transcription.
• DBPs play roles in chronic diseases and are targets for antibiotic and steroid production.
• Existing DBP identification methods require further enhancement for accuracy.

Purpose of the Study:

• To design and validate a novel, highly accurate predictor for identifying DNA-binding proteins.
• To improve upon existing methods for DBP identification using advanced feature engineering and machine learning.

Main Methods:

• Feature extraction using Filtered position-specific scoring matrix (F-PSSM), Position specific scoring matrix-dipeptide composition (PSSM-DPC), and Reduced position-specific scoring matrix (R-PSSM).
• Application of discrete wavelet transform (DWT) to F-PSSM, PSSM-DPC, and R-PSSM to create novel descriptors: F-PSSM-DWT, PSSM-DPC-DWT, and R-PSSM-DWT.
• Training and evaluation of machine learning models including Light eXtreme gradient boosting (LiXGB), eXtreme gradient boosting (XGB), extremely randomized trees (ERT), and Adaboost.

Main Results:

• The LiXGB model combined with the R-PSSM-DWT descriptor achieved superior performance.
• This novel predictor demonstrated a 6.55% increase in training accuracy and a 5.93% increase in testing accuracy compared to existing methods.
• The developed predictor, DBP-iDWT, significantly outperforms previous predictors.

Conclusions:

• The novel predictor exhibits excellent performance, surpassing existing methods for DBP identification.
• The DBP-iDWT predictor holds promise for developing more effective therapeutic strategies against fatal diseases.
• This research contributes to advancing DBP identification techniques in bioinformatics and molecular biology.