MiPepid

MiPepid predicts whether short open reading frames (sORFs) encode micropeptides, i.e., small proteins of 100 amino acids or fewer, using machine learning.

Key Features:

• Input: Accepts DNA fasta sequences and searches those sequences for candidate sORFs.
• sORF search: Identifies all sORFs up to 303 base pairs across all three translation frames.
• Feature extraction: Derives 4-mer features solely from ORF sequences for model input.
• Model: Employs logistic regression to predict coding potential of each sORF.
• Output: Assigns a predicted class label (coding or noncoding) with an associated probability and compiles results into an output .csv file.
• Training data: Trained on meticulously curated datasets derived from existing databases.
• Specialization: Tailored for micropeptide prediction and distinguished from coding potential tools optimized for longer proteins.
• Performance: Achieves 96% success rate on a blind dataset of high-confidence micropeptides and classifies newly discovered micropeptides not included in training or test sets.

Scientific Applications:

• Micropeptide discovery: Identification and prioritization of sORFs likely to encode bona fide micropeptides for downstream experimental validation.
• Genome and transcriptome annotation: Annotation of short coding sequences that are frequently overlooked by tools optimized for longer proteins.
• Benchmarking: Comparative evaluation of coding potential prediction methods focused on short ORFs.

Methodology:

Processes DNA fasta sequences to find sORFs up to 303 bp in all three translation frames, extracts 4-mer features from ORF sequences, trains a logistic regression model on curated datasets from existing databases, predicts coding versus noncoding labels with probabilities, and outputs results to a .csv file.