FOLDALIGN

FOLDALIGN performs structural pairwise alignment of RNA sequences using the Sankoff algorithm to detect conserved non-coding RNAs (ncRNAs) and other structural RNA elements (eleRNA) for gene finding and motif detection in genomic regions.

Key Features:

• Algorithm: Uses the Sankoff algorithm for simultaneous RNA sequence alignment and secondary-structure prediction.
• Alignment modes: Supports both local and global pairwise alignments and scanning for common structural motifs of limited size.
• Scoring scheme: Employs structural parameters akin to free energy calculations together with substitution matrices similar to RIBOSUM.
• Low-sequence-similarity detection: Capable of detecting conserved RNA structures when sequence similarity is below 40%.
• Performance metrics: Demonstrated structure prediction performance with a Matthews correlation coefficient of approximately 0.7, sensitivity ~0.8, and positive predictive value ~0.9.
• Multithreading and scalability: Multithreaded implementation improves execution time (up to ~5×) and enables comparison of longer sequences (approximately 2000–6000 nucleotides).
• Validation datasets: Validated on datasets where ncRNAs and eleRNAs are energetically indistinguishable from surrounding genomic context.

Scientific Applications:

• ncRNA and eleRNA detection: Identification of structurally conserved non-coding RNAs and other structural RNA elements that lack strong sequence conservation.
• Gene finding in genomic regions: Locating genes or RNA motifs with low sequence similarity within larger genomic contexts.
• Motif scanning and family localization: Scanning pairs of sequences for common structural motifs and locating RNA families within genomic contexts.
• Large-scale and long-sequence analyses: Enabling large-scale screens and comparisons of sequences up to several thousand nucleotides via multithreading.

Methodology:

Pairwise alignments are computed with the Sankoff algorithm for simultaneous alignment and folding with options for local or global alignment; scoring combines structural free-energy-like parameters and RIBOSUM-like substitution matrices; a multithreaded implementation accelerates execution and permits handling sequences of ~2000–6000 nucleotides.