SubFeat

SubFeat implements a feature-subspace ensemble classifier to predict functional properties of DNA, RNA, and protein/peptide sequences.

Key Features:

• Feature Representation: Generates multiple numerical feature representation schemes from primary structure sequences for DNA, RNA, and protein/peptide data.
• Customizable Feature Overlapping: Provides configurable feature overlapping options to control how features are grouped into subspaces.
• Ensemble Classification: Divides the feature space into subspaces, trains individual classifiers on each subspace, and combines base classifiers via a weighted majority voting mechanism.
• Machine Learning Support: Supports training and evaluation using various machine learning classifiers on the generated feature representations.
• Benchmark Validation: Validated on four datasets (two DNA, one RNA, one protein), demonstrating improved performance relative to single classifiers and other ensemble methods.

Scientific Applications:

• DNA function prediction: Predicts functional properties of DNA sequences from numerical feature representations.
• RNA function prediction: Predicts functional properties of RNA sequences using subspace-based ensemble models.
• Protein/peptide function prediction: Predicts functional properties of protein and peptide sequences from primary sequence features.
• Genomics and proteomics annotation: Supports functional annotation tasks in genomics and proteomics research through ensemble-based classification.

Methodology:

Generates numerical feature representations from primary sequences; partitions the feature space into (optionally overlapping) subspaces; trains separate classifiers on each subspace; aggregates base classifiers using a weighted majority voting scheme.