LGEpred

LGEpred predicts gene expression levels from protein sequence amino acid composition to infer relationships between sequence features and gene regulation.

Key Features:

• Correlation Analysis: LGEpred computes correlations between gene expression levels and amino acid compositions, identifying positive correlations for residues such as Ala, Gly, Arg, Val and negative correlations for residues such as Asp, Leu, Asn, Ser.
• Support Vector Machine (SVM) Prediction: The method employs a Support Vector Machine trained on known expression data to predict gene expression from protein sequences within the same organism under similar conditions.
• Dipeptide Composition Enhancement: Incorporation of dipeptide composition into the predictive model increases performance, improving correlation coefficients from 0.70 to 0.72 between predicted and experimentally determined expressions.
• Cross-Validation Testing: Model robustness is assessed using 5-fold cross-validation.
• Functional Classification Improvement: Amino acid composition and predicted expression are used to refine protein function classification by linking sequence features with functional outcomes.

Scientific Applications:

• Gene Expression Prediction: Predicts gene expression levels from protein sequences to infer regulatory patterns without direct experimental measurement.
• Protein Function Classification: Uses sequence composition and expression data to assist classification of protein function.
• Evolutionary Studies: Enables analysis of evolutionary trends by integrating gene expression data with sequence composition to study co-evolution of genetic and proteomic features.

Methodology:

Analysis of 3468 genes from Saccharomyces cerevisiae with computation of correlations between amino acid composition and gene expression, training of a Support Vector Machine on known expression data, incorporation of dipeptide composition into the model, and validation by 5-fold cross-validation.