TDimpute

TDimpute imputes missing gene expression (RNA-seq) values from DNA methylation profiles using a transfer learning-based neural network to enable downstream analyses such as identification of methylation-driven genes and survival analysis.

Key Features:

• Methylation-guided imputation: Uses DNA methylation profiles as the input source to infer missing RNA-seq gene expression values.
• Transfer learning framework: Trains a general model on a pan-cancer dataset from The Cancer Genome Atlas (TCGA) to capture broad cross-cancer patterns.
• Neural network architecture: Implements a neural network-based model to learn relationships between methylation and gene expression.
• Fine-tuning for specificity: Fine-tunes the pretrained pan-cancer model on specific target cancer datasets to improve cancer-type-specific imputation accuracy.
• Performance superiority: Demonstrates 7%–11% improvement in imputation accuracy over state-of-the-art methods across 16 diverse cancer datasets under varying missing-data rates.
• Generalizability: Validated on an independent Wilms tumor dataset from the TARGET project, indicating applicability beyond TCGA datasets.

Scientific Applications:

• Imputation of missing RNA-seq data: Reconstructs missing gene expression profiles for samples with incomplete RNA-seq measurements.
• Identification of methylation-driven genes: Enables detection of genes whose expression is driven by DNA methylation changes.
• Prognostic marker discovery: Supports identification of genes related to prognosis from imputed expression data.
• Clustering and subtype analysis: Facilitates clustering analyses and cancer subtype characterization using reconstructed expression profiles.
• Survival analysis: Permits survival analyses within TCGA and other datasets using imputed gene expression.

Methodology:

TDimpute employs a transfer learning-based neural network trained on a pan-cancer TCGA dataset and subsequently fine-tuned on specific target cancer datasets, using DNA methylation profiles as input to impute gene expression.