Prediction of donor splice sites using random forest with a new sequence encoding approach

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate splice site detection is crucial for predicting gene structure.
• Developing efficient splice site prediction methods is vital for biological research.

Purpose of the Study:

• To present a novel sequence encoding approach for donor splice site prediction.
• To evaluate the performance of various machine learning classifiers using this approach.

Main Methods:

• Encoding donor splice site motifs into numeric vectors based on adjacent di-nucleotide dependencies.
• Utilizing Random Forest (RF), Support Vector Machines (SVM), Artificial Neural Network (ANN), and other classifiers.
• Evaluating performance on Homo sapiens donor splice site data from the HS3D dataset.

Main Results:

• The Random Forest classifier outperformed all other considered classifiers.
• RF achieved higher prediction accuracy than existing methods like MEM, MDD, WMM, MM1, NNSplice, and SpliceView on an independent test dataset.

Conclusions:

• An online prediction server, MaLDoSS, was developed for predicting donor splice sites.
• The proposed approach offers computational feasibility and high prediction accuracy for eukaryotic gene structure analysis.