An Ensemble Tf-Idf Based Approach to Protein Function Prediction via Sequence Segmentation.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Modeling variable-length protein sequences is crucial for accurate function prediction.
• Existing tf-idf based methods face challenges with sequence length variations.
• There is a need for robust feature sets that handle diverse protein sequence lengths effectively.

Purpose of the Study:

• To develop an efficient method for modeling variable-length protein sequences using tf-idf variants.
• To enhance protein function prediction accuracy and consistency.
• To introduce ProtVecGen-Ensemble, an ensemble model for improved feature representation.

Main Methods:

• Utilized length-normalized-tf-idf and log-normalized-tf-idf descriptors.
• Employed a segmentation technique for processing protein sequences.
• Developed ProtVecGen-Ensemble, an ensemble of three models trained on segmented datasets.
• Evaluated performance on biological process (BP) and molecular function (MF) datasets.

Main Results:

• ProtVecGen-Ensemble demonstrated superiority over contemporary tf-idf based methods.
• Achieved significant improvements: +6.07% (BP) and +7.56% (MF) over MLDA.
• Combined model (ProtVecGen-Ensemble + ProtVecGen-Plus) yielded further gains: +8.90% (BP) and +11.28% (MF) over MLDA.
• Showcased enhanced performance consistency with a novel variance-based metric, reducing variance by up to 56.85% (BP) and 56.08% (MF).

Conclusions:

• The proposed tf-idf based feature set and ProtVecGen-Ensemble model offer superior performance for protein function prediction.
• The method effectively addresses the challenge of variable protein sequence lengths, improving consistency.
• The combined framework represents a significant advancement in computational protein analysis and prediction.