SWAM

SWAM integrates multiple transcriptome imputation models using summary-level data to improve gene expression imputation accuracy for transcriptome-wide association studies (TWAS).

Key Features:

• Multi-model integration: Integrates an arbitrary number of transcriptome imputation models across tissues and datasets.
• Summary-level operation: Combines imputed expression levels using summary-level data without requiring individual-level genotype or expression data.
• Linear optimization: Linearly optimizes weights to combine imputed expression from multiple models to maximize imputation accuracy for a target tissue.
• eQTL and sample-size leverage: Exploits shared expression quantitative trait loci (eQTLs) across tissues and increased effective sample size to improve accuracy.
• Meta-TWAS capability: Extends meta-imputation to enable meta-TWAS by integrating multiple tissues in TWAS at the summary level.
• Empirical validation: Validated by combining 49 tissue-specific models from the GTEx Project and the Depression Susceptibility Genes and Networks (DGN) Project and evaluated in GEUVADIS lymphoblast cell lines samples with superior accuracy relative to single-tissue and existing multi-tissue methods.

Scientific Applications:

• Transcriptome-wide association studies (TWAS): Improves power and accuracy of TWAS by enhancing gene expression imputation for target tissues.
• Meta-TWAS analyses: Enables integration of multiple tissues at the summary level to detect regulatory effects across tissues.
• GWAS interpretation: Facilitates interpretation of genome-wide association study (GWAS) signals through more accurate inferred gene expression.
• Mendelian Randomization analyses: Supports Mendelian Randomization by providing improved imputed expression instruments derived from multiple models.

Methodology:

Linearly optimizes weights to combine an arbitrary number of transcriptome imputation models using summary-level data to maximize imputation accuracy for a target tissue.