mLoc-mRNA

mLoc-mRNA predicts sub-cellular localizations of mRNAs from sequence-derived k-mer features using machine learning.

Key Features:

• Feature Transformation: Transforms each mRNA sequence into a numeric feature vector of size 5460 using k-mer features with k = 1–6.
• Feature Selection: Uses an Elastic Net statistical model to select 1812 informative features from the initial 5460 k-mer features.
• Localization Prediction: Applies a Random Forest supervised learning algorithm on the selected features to predict mRNA sub-cellular localizations.
• Predicted Localizations: Predicts nine localization categories: cytoplasm, cytosol, endoplasmic reticulum, exosome, mitochondrion, nucleus, pseudopodium, posterior, and ribosome.
• Performance Evaluation: Validated with five-fold cross-validation (accuracies: 70.87%, 68.32%, 68.36%, 68.79%, 96.46%, 73.44%, 70.94%, 97.42%, 71.77% for cytoplasm, cytosol, endoplasmic reticulum, exosome, mitochondrion, nucleus, pseudopodium, posterior, and ribosome, respectively) and an independent test set (accuracies: 65.33%, 73.37%, 75.86%, 72.99%, 94.26%, 70.91%, 65.53%, 93.60%, 73.45% for the same localizations).

Scientific Applications:

• mRNA localization prediction: Enables computational prediction of mRNA localization patterns across multiple sub-cellular compartments.
• Complement experimental methods: Provides in silico localization hypotheses to complement experimental localization assays.
• Cellular distribution and gene regulation studies: Supports investigation of mRNA distribution within cells and its implications for gene expression regulation.

Methodology:

Convert mRNA sequences to 5460-dimensional k-mer feature vectors (k=1–6), apply Elastic Net to select 1812 features, train a Random Forest classifier for localization prediction, and validate using five-fold cross-validation and an independent test set.