csORF-finder

csORF-finder identifies coding short open reading frames (csORFs) across multiple species, including Homo sapiens, Mus musculus, and Drosophila melanogaster, to detect sORFs (≤303 nucleotides) in mRNAs and long non-coding RNAs (lncRNAs) that may encode functional peptides.

Key Features:

• Ensemble Learning Framework: Employs an ensemble learning approach integrating Efficient-CapsNet with LightGBM to distinguish coding sORFs from non-coding sequences.
• Trinucleotide Deviation from Expected Mean (TDE): Introduces the TDE feature encoding scheme to capture trinucleotide composition deviations for improved csORF prediction.
• In-frame Sequence-based Features: Implements i-framed-3mer, i-framed-CKSNAP, and i-framed-TDE to represent in-frame sequence characteristics of candidate sORFs.
• Comparison to Traditional Features: Demonstrates improved performance over traditional 3-mer, CKSNAP, and original TDE feature representations.
• Multi-Species Capability: Designed for cross-species analysis including Homo sapiens, Mus musculus, and Drosophila melanogaster.
• Non-ATG Initiation Independence: Detects csORFs without reliance on canonical ATG start codons, enabling identification of non-ATG-initiated peptides.

Scientific Applications:

• Functional Peptide Discovery: Enables identification of novel peptide-encoding sORFs for downstream experimental validation.
• lncRNA Analysis: Screens lncRNA datasets to detect potential csORFs within long non-coding RNAs.
• High-Throughput Screening: Supports large-scale identification of csORFs across genomes for comparative and functional studies.

Methodology:

Uses an ensemble of Efficient-CapsNet and LightGBM trained on features including trinucleotide deviation from expected mean (TDE), i-framed-3mer, i-framed-CKSNAP, and i-framed-TDE, and reports performance on multi-species and non-ATG initiation independent test datasets.