PALMA

PALMA predicts optimal local alignments between mRNA/EST sequences and genomic DNA by combining splice-site prediction with a large-margin learning framework to identify intron boundaries for transcriptome and genomic analyses.

Key Features:

• Large Margin Learning Framework: Employs a large margin learning framework that optimizes model parameters by solving a convex optimization problem so true alignments are scored higher than incorrect ones.
• Accurate Splice Site Prediction: Predicts splice sites precisely to identify intron boundaries and achieve optimal local alignment between mRNA/EST sequences and genomic DNA.
• Robustness to Genetic Variations: Maintains high accuracy in the presence of genetic variations including mutations, insertions, deletions, and substantial noise.
• Superior Experimental Performance: Correctly identified intron boundaries in 5702 artificially shortened EST sequences from Caenorhabditis elegans and humans with only two exceptions in an experiment involving artificially generated micro-exons, outperforming exalin which misaligned 37 sequences.

Scientific Applications:

• Gene expression analysis: Provides precise mRNA-to-genome alignments to support transcript structure determination and expression studies.
• Transcriptomics: Facilitates accurate mapping of EST/mRNA data to genomic loci for transcript discovery and annotation.
• Genomics: Aids in defining exon–intron structures and refining genome annotations.
• Alternative splicing and micro-exon research: Enables detection and characterization of alternative splicing events and micro-exons through accurate intron boundary identification.
• Downstream analyses (variant calling, functional annotation, evolutionary studies): Improves input alignments for variant calling, functional annotation of transcripts, and comparative/evolutionary analyses.

Methodology:

Integrates accurate splice-site predictions with sequence alignment techniques and tunes model parameters using a large-margin learning approach implemented via convex optimization to prioritize true alignments.