MLC

MLC learns GO-term-specific sample weights to compute weighted co-expression measures from RNA-Seq data for improved gene function prediction while mitigating batch effects and irrelevant experimental conditions.

Key Features:

• GO-Term-Specific Weighting: Assigns weights to expression samples based on their relevance to individual Gene Ontology (GO) terms.
• Weighted Co-Expression Measure: Computes a weighted co-expression metric per GO term as an alternative to unweighted Pearson correlation.
• Maximization and Minimization Strategy: Maximizes weighted co-expression for gene pairs co-annotated with a GO term and minimizes it when one gene lacks that annotation.
• Dynamic Sample-Weight Optimization: Uses a fast algorithm that dynamically adjusts sample weights to optimize term-specific co-expression measures.
• Batch-Effect and Irrelevant-Condition Mitigation: Down-weights noninformative samples to reduce the influence of batch effects and irrelevant experimental conditions on co-expression.
• Improved Term-Centric Performance: Enhances Guilt-By-Association gene function prediction performance relative to Pearson correlation, demonstrated on Arabidopsis thaliana RNA-Seq data.

Scientific Applications:

• Functional Genomics: Refining gene function prediction and annotation transfer using GO-term-specific co-expression from RNA-Seq data.
• Biological Process Gene Prioritization: Prioritizing genes associated with specific biological processes defined by Gene Ontology terms.
• Transcriptomic Analysis in Model Organisms: Applying term-centric co-expression analysis to RNA-Seq datasets such as Arabidopsis thaliana to improve functional inference.

Methodology:

MLC optimizes GO-term-specific sample weights with a fast algorithm to compute weighted co-expression measures, applying maximization for co-annotated gene pairs and minimization when one gene lacks the annotation, and compares weighted co-expression against unweighted Pearson correlation; experiments include Arabidopsis thaliana RNA-Seq data.