GeneMark

GeneMark predicts protein-coding genes in fungal genomes using an ab initio hidden Markov model framework with iterative unsupervised self-training and an enhanced intron submodel accommodating branch point variability across Ascomycota, Basidiomycota, and Zygomycota.

Key Features:

• Ab Initio Algorithm: Operates without external predefined training sets by estimating model parameters directly from the input genome.
• Iterative Unsupervised Training: Performs iterative unsupervised training on the input genomic sequence to estimate HMM parameters.
• Enhanced Intron Submodel: Includes an intron submodel that accommodates sequences with and without branch point sites to handle splicing variation across fungal phyla.
• Self-Training Mechanism: Applies a multi-step self-training procedure that incrementally increases intron submodel complexity during training.
• High Accuracy: Demonstrates improved accuracy for exon and whole-gene structure prediction compared with gene finders that rely on supervised training.

Scientific Applications:

• Fungal genome annotation: Identifies and annotates protein-coding genes in fungal genomes without requiring comprehensive external training sets.
• Genome sequencing projects: Streamlines annotation of newly sequenced fungal genomes by deriving model parameters directly from the sequence data.
• Comparative and evolutionary analyses: Provides accurate gene models that support inference of gene function and evolutionary relationships within fungal species.

Methodology:

Uses iterative unsupervised self-training to estimate hidden Markov model parameters from the input genome sequence and progressively refines an intron submodel by increasing its complexity to capture branch point variability.