WebAUGUSTUS

WebAUGUSTUS predicts protein-coding genes and alternative splice variants in eukaryotic genomic sequences and supports parameter training and constraint-guided gene model refinement for genome annotation.

Key Features:

• Ab initio gene prediction: Uses a generalized Hidden Markov Model (HMM) to perform de novo prediction of protein-coding genes from eukaryotic genomic sequences.
• Intron and splice-site modeling: Implements novel intron length distribution modeling and a donor splice site model to improve accuracy on longer genomic regions and multi-gene sequences.
• User-defined constraints: Incorporates positional constraints for splice sites, translation initiation sites, stop codons, known exons, and exon/intron intervals to include partial gene-structure evidence such as expressed sequence tags (ESTs) or protein alignments.
• Multiple transcript prediction: Predicts multiple splice variants per gene to represent alternative splicing.
• Motif searching: Supports searching user-defined regular expressions against putative proteins encoded by predicted genes.
• Training capabilities: Trains AUGUSTUS parameters from genomic sequences together with expressed sequence tags or protein sequences to produce species-optimized parameter sets.
• GC-content dependent parameter estimation: Estimates parameters dependent on GC-content to adapt predictions to regional compositional variation.

Scientific Applications:

• Eukaryotic genome annotation: Annotates newly sequenced and assembled eukaryotic genomes by predicting protein-coding genes.
• Alternative splicing analysis: Characterizes alternative splice variants and transcript diversity within gene loci.
• Evidence-guided gene model refinement: Refines gene models using ESTs, protein alignments, and user-specified positional constraints.
• Species-specific parameter optimization: Produces and applies trained AUGUSTUS parameter sets for improved predictions on specific genomes.

Methodology:

Applies a generalized Hidden Markov Model (HMM) integrating submodels including intron length distribution and a donor splice site model, uses GC-content dependent parameter estimation, and supports parameter training from genomic sequences with expressed sequence tags or protein sequences.