Major updates in M-cTWAS software
- Extended the cTWAS framework to incorporate prediction models from multiple molecular trait types across diverse biological contexts.
- Redesigned the software interface to improve modularity, implementing key M-cTWAS tasks as standalone components. Users may run the full M-cTWAS analysis via the main function or execute individual steps independently.
- Implemented a “no-LD” mode that enables M-cTWAS analyses without requiring an LD reference panel.
Other important updates
Input processing
- Supports multiple prediction model formats, including PredictDB, FUSION, and top-QTL.
- Provides streamlined preprocessing tools to harmonize GWAS summary statistics, prediction models, and LD reference.
- Offers utility functions for constructing LD matrices from individual-level genotype data and for loading LD matrices of various formats, with flexible support for user-defined custom formats.
Computation
- Introduces new parameter estimation procedures that enable sharing of prior effect variance parameters across multiple groups of molecular traits. Several schemes of parameter sharing are supported, including: shared variance for all groups of variables (including SNPs); shared variance for all molecular traits; shared variance for all molecular traits of the same type/modality.
- Compute standard errors and p-values of enrichment parameters.
- Implements new methods to screen genomic regions with likely causal molecular trait signals. This speeds up the computation, as only such regions will be subject to full fine-mapping (allowing more than one causal signal).
- Supports fine-mapping of single or multiple regions, with or without LD information.
- Re-implements a simplified single effect regression (SER) model from SuSiE, incorporating the new implementation of null configuration. It is used in EM to speed up computation.
- Improves computational efficiency by more than ten-fold, enabling large-scale analyses across many molecular traits. A typical M-cTWAS analysis using molQTLs from multiple groups could be completed within 1–3 hours.
Creating the output
- Introduces new methods to combine evidence from all molecular traits targeting the same genes.
- Enhances visualization of M-cTWAS results, including updated locus plots with integrated gene tracks and fine-mapping result panels.
- Incorporates LD-mismatch diagnosis from SuSiE-RSS in post-processing.
- Introduces a region-merging step to resolve the “cross-region” challenge, where the genetic variants contributing to a molecular trait spans two regions.
Note: this updated cTWAS version currently only works for GWAS summary statistics. To use individual level data, please refer to the old cTWAS version for now. We will release a new version that will work with individual level data later.