Gene-Bench

Gene-Bench benchmarks algorithms for detecting differentially expressed genes to enable comparative evaluation of gene expression analysis methods.

Key Features:

• Benchmarking Framework: Provides an integrated system to evaluate the efficacy and comparative performance of differential gene detection algorithms.
• MIDGET (Machine learning Identification Differential Gene Expression Tool): Implements machine learning approaches including extreme gradient boosting and deep neural networks for differential expression detection.
• Comprehensive Dataset: Includes 73 transcription-factor perturbation experiments validated by Chip-seq data, 129 drug perturbation experiments, and synthetic data generated through proprietary methods.
• Evaluation Metrics: Employs Kolmogorov, F1 score, and AUC/ROC to assess algorithm performance from multiple perspectives.
• Algorithm Flexibility: Implemented in Python with support for integration of algorithms developed in R.

Scientific Applications:

• Benchmarking differential gene detection: Compare and quantify performance of statistical and machine learning methods for detecting differentially expressed genes.
• Genome-wide analysis and expression profiling: Evaluate methods for genome-wide expression studies using transcription-factor perturbation and drug perturbation datasets.
• Method comparison across data types: Assess algorithm robustness on real Chip-seq-validated experiments and synthetic datasets.

Methodology:

Uses real data (73 transcription-factor perturbation experiments validated by Chip-seq and 129 drug perturbation experiments) and synthetic data to benchmark algorithms, applies MIDGET using extreme gradient boosting and deep neural networks, and evaluates performance with Kolmogorov, F1 score, and AUC/ROC; implemented in Python with support for R algorithms.