SpliceVec-g

SpliceVec-g predicts splice junctions from genomic and intronic sequences using distributed feature representations (SpliceVec) and a multilayer perceptron (MLP) to improve identification of canonical and non-canonical splice sites and support analysis of alternative splicing.

Key Features:

• Distributed Feature Representation: Employs a distributed feature representation model (SpliceVec) inspired by natural language processing to avoid manual feature extraction biases.
• Machine Learning Integration: Uses a multilayer perceptron (MLP) to classify splice junctions based on the learned SpliceVec representations.
• Comprehensive Contextual Analysis: Incorporates entire intronic sequences rather than only flanking regions to improve feature extraction and prediction accuracy.
• Canonical and Non-Canonical Junction Detection: Designed to be invariant to both canonical and non-canonical splice junctions.
• Robustness in Varied Data Conditions: Maintains high performance on reduced and class-imbalanced datasets.
• Computational Efficiency: Enables rapid training and application using user-defined datasets.

Scientific Applications:

• Gene structure and alternative splicing analysis: Facilitates delineation of gene structure and investigation of functional diversity imparted by alternative splicing.
• RNA-seq intron boundary identification: Addresses challenges associated with traditional RNA-seq-based identification of intron boundaries.
• Large-scale genomic studies: Supports studies on large or limited/imbalanced datasets for splice junction discovery.

Methodology:

Generates distributed feature representations (SpliceVec) from entire intronic sequences inspired by NLP approaches and classifies splice junctions using a multilayer perceptron (MLP); training and evaluation are performed on user-defined datasets, including reduced and class-imbalanced datasets.