BPLA kernel

BPLA kernel applies a profile-based base-pairing local alignment (BPLA) kernel to multiple-sequence alignments and uses Support Vector Machines (SVMs) to predict and classify noncoding RNAs (ncRNAs).

Key Features:

• Integration of Alignment Data: Uses profile information from multiple sequence alignments to capture evolutionary relationships and structural similarities among RNA sequences.
• Support Vector Machines (SVMs): Combines the BPLA kernel with Support Vector Machines for ncRNA prediction and classification.
• Robustness Against Alignment Errors: Maintains high performance when input alignments contain errors, including (a) sequences aligned without considering secondary structures and (b) alignments containing unrelated non-ncRNA sequences.
• Improved Accuracy: Outperforms other profile-based prediction methods in benchmark tests using high-quality structural alignment datasets.

Scientific Applications:

• Family Prediction: Predicts and classifies ncRNAs into families by assessing structural and profile-based similarities.
• Hierarchical Clustering: Supports organization of ncRNAs based on evolutionary relationships and structural features for clustering and phylogenetic analyses.
• Remote Homology Search: Detects distant homologs to identify novel ncRNA sequences with functional relevance across species.

Methodology:

Applies a profile-based BPLA kernel to alignment profiles and trains Support Vector Machines (SVMs) for classification; performance was assessed by benchmark testing on high-quality structural alignments and by robustness tests against the two specified alignment-error types.