NetActivity

NetActivity computes gene set activity scores from gene expression data using sparsely-connected autoencoders trained on biological pathway annotations.

Key Features:

• Gene Set Score Computation: Computes gene set scores using the `computeGeneSetScores` function with pre-trained or custom sparsely-connected autoencoder models.
• Data Preparation: Prepares input data via the `prepareSummarizedExperiment` function for compatibility with the scoring framework.
• Pre-trained Models: Provides access to pre-trained models through the companion `NetActivityData` package, including models trained on 1518 GO biological processes terms and KEGG pathways using all GTEx samples.
• Robust Score Generation: Generates gene set activity scores robust to initialization parameters and representative of the original transcriptome through a three-tier training approach.
• Interpretable Architecture: Employs an autoencoder where each neuron in the inner layer represents a gene set, assigning higher importance to more biologically relevant genes.
• Cross-platform Integration: Enables combining bulk RNA-seq and microarray datasets for meta-analysis.

Scientific Applications:

• Cancer Progression Analysis: Identifies gene sets related to cell division in prostate cancer progression through meta-analysis of combined bulk RNA-seq and microarray datasets.
• Drug Resistance Studies: Detects alterations in gene sets associated with cancer progression due to drug resistance in metastatic prostate cancer.
• Comparative Pathway Analysis: Returns gene set activity scores with more consistent definition and higher interpretability than GSVA and hipathia for pathway-based transcriptome interpretation.

Methodology:

NetActivity uses a sparsely-connected autoencoder with each neuron in the inner layer representing a gene set, trained through a three-tier training procedure on 1518 GO biological processes terms and KEGG pathways using all GTEx samples.