egpred

egpred integrates BLASTX and BLASTN similarity searches with ab initio predictions and NNSPLICE-based splice-site reassignment to improve exon-level gene prediction accuracy.

Key Features:

• Hybrid approach: Combines ab initio predictions with similarity searches to balance specificity and sensitivity for exon detection.
• Initial BLASTX search: Searches the genomic sequence against the RefSeq database with E-value < 1 to identify coding region hits.
• Relaxed BLASTX search: Uses initial hits as queries with relaxed parameters (E-value < 10) to capture additional exon candidates.
• BLASTN intron detection: Searches the genomic sequence against an intron database to identify probable intron regions.
• Exon–intron comparison and filtering: Compares predicted exon and intron regions to filter out incorrect exon predictions.
• Splice-site reassignment: Uses NNSPLICE to reassess and reassign splicing signal positions within probable coding exons.
• Integration with ab initio predictions: Integrates refined exon data with ab initio outputs considering start/stop and splice signal strengths.
• Benchmark performance: Reports a 4%–10% performance improvement on HMR195 and Burset/Guigo datasets.
• Large-scale demonstration: Applied to approximately 95 Mbp of human chromosome 13 for large-fragment gene prediction.
• Computational intensity: Multiple BLAST runs required per analysis increase computational demand.

Scientific Applications:

• Exon-level gene prediction: Identification and refinement of exon regions within genomic sequences.
• Intron detection and boundary refinement: Detection of probable introns and refinement of exon–intron boundaries using BLASTN and NNSPLICE.
• Benchmarking ab initio methods: Enhancing and evaluating ab initio program performance on benchmark datasets such as HMR195 and Burset/Guigo.
• Large genomic region analysis: Gene prediction on large genomic fragments, exemplified by a ~95 Mbp region of human chromosome 13.

Methodology:

Sequential computational steps: initial BLASTX of the genomic sequence against RefSeq (E-value < 1); relaxed BLASTX using initial hits as queries (E-value < 10); BLASTN against an intron database; comparison and filtering of probable exon and intron regions; reassignment of splicing signals with NNSPLICE; and integration of refined exon data with ab initio predictions using start/stop and splice-site strength metrics.