MED

MED predicts prokaryotic genes using entropy-based statistical models to improve annotation of open reading frames (ORFs) and translation initiation sites (TISs) for comparative genomic analyses.

Key Features:

• Non‑supervised gene prediction: Implements an iterative self-learning process to derive genome-specific parameters without supervised training.
• TriTISA (Translation Initiation Site Self-learning Algorithm): Applies a self-learning algorithm specifically to improve prediction accuracies of 5′ and 3′ gene ends and TIS locations.
• Entropy Density Profile (EDP) for ORFs: Represents protein-coding ORFs using a linguistic Entropy Density Profile to capture coding-sequence patterns.
• Multivariate Entropy Distance (MED) algorithm / MED 2.0: Integrates EDP and TIS-related features into a multivariate entropy distance framework implemented in MED 2.0.
• TIS-related feature set: Incorporates multiple features pertinent to translation initiation sites to refine TIS prediction.
• Performance in GC-rich and archaeal genomes: Demonstrates high prediction performance specifically noted for GC-rich and archaeal genomes.
• Comparative and annotation benchmarking: Shows superior 5′ and 3′ end match accuracy relative to existing gene finders and reveals differences from GenBank annotations.

Scientific Applications:

• Prokaryotic gene annotation: Predicts protein-coding genes and refines ORF boundaries in prokaryotic genomes.
• Translation initiation site mapping: Improves localization of TISs and characterization of 5′ and 3′ gene ends.
• Comparative genomics: Enables comparative analyses of gene structures across diverse prokaryotic genomes.
• Archaeal translation initiation studies: Detects divergent translation initiation mechanisms in archaeal genomes for mechanistic investigation.
• Annotation improvement in GC-rich genomes: Provides enhanced gene boundary annotation in GC-rich prokaryotic genomes.

Methodology:

Uses a non‑supervised, iterative self-learning approach combining a linguistic Entropy Density Profile for ORFs, TIS-related features, the TriTISA algorithm for TIS self-learning, and integration via the Multivariate Entropy Distance (MED) framework (MED 2.0).