LncCat

LncCat identifies long non-coding RNAs (lncRNAs) by integrating fused sequence information and an ensemble learning strategy to distinguish lncRNAs from protein-coding transcripts, including cases with short open reading frames (sORFs).

Key Features:

• CatBoost ensemble learning: Employs the CatBoost category boosting algorithm to construct a predictive model for lncRNA identification.
• ORF-attention features: Incorporates ORF-attention features that exhibit significant differences between lncRNAs and protein-coding transcripts.
• Fused sequence information: Integrates fused sequence information to comprehensively encode transcript sequences.
• Five feature types: Uses five distinct types of features to encode transcripts for model training.
• sORF and long ORF handling: Applies feature encoding and modeling approaches that address both short open reading frames (sORFs) and long ORFs.
• Visualization comparisons: Utilizes visualization comparisons to demonstrate feature differences between lncRNAs and protein-coding transcripts.

Scientific Applications:

• lncRNA identification: Distinguishes lncRNAs from protein-coding transcripts, accounting for the presence of sORFs.
• Cross-species benchmarking: Evaluated on benchmark datasets with reported Matthew's Correlation Coefficient (MCC) scores of 0.9503 (human), 0.9219 (mouse), 0.8591 (zebrafish), 0.8672 (wheat), and 0.9047 (chicken), representing improvements ranging from 1.90% to 51.64% across datasets.
• sORF-specific improvement: Demonstrates MCC improvements on sORF test datasets of at least 11.90% (human), 12.96% (mouse), and 42.61% (zebrafish).

Methodology:

Integrates five types of sequence-derived features and fused sequence information, incorporates ORF-attention features, applies CatBoost ensemble learning to build a predictive model, uses visualization comparisons, and benchmarks performance on multiple species datasets.