ProsmORF-pred

ProsmORF-pred predicts and functionally annotates small open reading frames (smORFs) in prokaryotic genomes using machine learning to identify translation initiation sites and protein-like short sequences.

Key Features:

• Machine learning prediction: Employs machine learning methodologies to improve accuracy of smORF identification in prokaryotic genomic data.
• Initiation Site Recognition Model: Focuses on nucleotide sequences upstream of potential start codons and is trained using longer open reading frames (>100 amino acids) from the same genomes.
• Functional Protein Sequence Identification Model: Analyzes translated amino acid sequences and is trained on annotated smORFs from Escherichia coli to recognize protein-like short sequences.
• Benchmarking and performance: Was benchmarked against contemporary approaches using an annotated set of smORFs from 32 prokaryotic genomes and matches or surpasses state-of-the-art, particularly for smORFs of 10–30 amino acids.
• Functional annotation via ProsmORFDB: Performs sequence similarity and genomic neighborhood similarity searches against ProsmORFDB to provide functional context for predicted smORFs.

Scientific Applications:

• smORF discovery: Identification of small open reading frames in prokaryotic genome assemblies and annotations.
• Functional inference: Assignment of potential functions to predicted smORFs through sequence similarity and genomic neighborhood comparisons against ProsmORFDB.
• Short-protein characterization: Detection and study of very short proteins (10–30 amino acids) relevant to prokaryotic gene regulation and cellular processes.

Methodology:

Two machine learning models are used: an initiation site recognition model trained on nucleotide sequences upstream of start codons from ORFs >100 amino acids in the same genomes, and a functional protein sequence identification model trained on translated amino acid sequences of annotated smORFs from Escherichia coli; performance was benchmarked using an annotated set of smORFs from 32 prokaryotic genomes, and functional annotation is performed via sequence similarity and genomic neighborhood similarity searches against ProsmORFDB.