blockclust

BlockClust detects transcripts with similar processing patterns by encoding expression profiles in compact discrete structures and processing them with fast graph-kernel techniques for unsupervised clustering and family-specific discriminative modeling.

Key Features:

• Novel encoding of expression profiles: Transforms expression profiles into compact discrete structures that capture processing patterns without relying on sequence or secondary structure information.
• Fast graph-kernel techniques: Processes encoded expression profiles using efficient graph-kernel methods for computational scalability.
• Unsupervised clustering: Groups transcripts with similar processing patterns without prior functional annotation.
• Family-specific discriminative models: Develops targeted classification models for specific RNA families.
• Cross-context robustness: Maintains accuracy and robustness across different organisms, tissues and cell lines.

Scientific Applications:

• Non-coding RNA functional annotation: Assigns putative function to unannotated ncRNAs by clustering transcripts according to their processing patterns in small RNA-seq reads data.
• Detection of RNA classes: Identifies distinct RNA classes based on patterns of processing and their traces in small RNA-seq data, circumventing limitations of sequence-based and secondary structure-based approaches.
• Analysis of post-transcriptionally modified transcripts: Enables classification of RNAs whose sequence information is altered by post-transcriptional modifications.

Methodology:

BlockClust encodes expression profiles in compact discrete structures and processes them using fast graph-kernel techniques for unsupervised clustering and family-specific discriminative modeling.