ANGLE

ANGLE predicts coding sequences from low-quality complementary DNA (cDNA) using a machine-learning framework that tolerates sequencing errors and incomplete data.

Key Features:

• Error-tolerant prediction: Handles sequencing artifacts and common inaccuracies such as frame-shifts and substitutions to maintain coding-sequence detection in unfinished cDNAs.
• Machine-learning inference: Employs a machine-learning approach to evaluate coding potential from limited or error-prone sequence segments.
• Codon usage and protein-structure integration: Integrates codon usage patterns and protein structural information to assess coding potential beyond stochastic-model approaches.
• Sequence-length independence: Maintains predictive accuracy independent of input sequence length, including short fragments.

Scientific Applications:

• Preliminary gene analysis: Predicts protein-coding regions from low-quality cDNA to support initial gene analysis.
• Comparative performance evaluation: Compared to ESTSCAN, ANGLE improved average Matthews's correlation coefficient by 9.26% on short sequence datasets (<1000 bases) and achieved comparable results on longer sequences.
• Gene discovery and annotation: Supports gene discovery and annotation from incomplete or error-prone cDNA data.
• Functional genomics: Facilitates functional genomics studies when high-quality cDNA is unavailable.

Methodology:

ANGLE applies a machine-learning framework that integrates codon usage patterns and protein structural data to evaluate coding potential in error-prone or incomplete cDNA sequences.