plotMI

plotMI computes and visualizes pairwise mutual information between positional k-mer distributions to reveal dependencies within DNA, RNA, and amino acid sequences and to aid interpretation of machine learning models, including deep learning, trained on sequence data.

Key Features:

• Mutual Information Calculation: Computes pairwise mutual information between positions using k-mer distributions to quantify dependencies and interactions within sequence datasets.
• Model Interpretation: Analyzes dependencies learned by machine learning and deep learning models to reveal positional preferences and pairwise interactions reflected in model outputs.
• Comparative Analysis: Compares mutual information patterns between different models and between model-derived and training-data distributions to assess how models capture sequence relationships.
• Visualization of Interactions: Produces visual representations of pairwise mutual information to depict positional k-mer dependencies across sequences.

Scientific Applications:

• Model Validation: Validates whether models capture dependencies present in training sequence data by comparing mutual information profiles.
• Comparative Studies: Compares modeling approaches to determine which better capture positional and pairwise sequence dependencies.
• Biological Insight Discovery: Identifies biologically relevant positional preferences and interactions within DNA, RNA, or amino acid sequences from mutual information patterns.

Methodology:

Calculates pairwise mutual information from positional k-mer distributions and visualizes the resulting pairwise dependencies; applies these calculations to both training data and model-derived sequence outputs for comparison.