Software

Everything below is open source — mostly C++ with MPI and CUDA for the inference engines, and Python for analysis and tooling.

• gVAMP

  Vector approximate message passing for GWAS. Joint inference of genetic effects with a spike-and-slab prior, p-value testing via state evolution, and out-of-sample prediction — scaled to 17 million whole-genome sequence variants on UK Biobank.

  • C++
  • MPI
  • HPC
  Repository Tutorial Paper

• TLgVAMP

  Transfer-learning gVAMP. Borrows strength from large, well-powered cohorts to improve polygenic risk score accuracy in statistically under-represented populations, while keeping population-specific signal identifiable.

  • C++
  • Transfer learning
  • Cross-ancestry PRS
  Repository

• Time2EVAMP (vampW)

  Approximate message passing for time-to-event data under a Weibull model. Joint variable selection over proteomic biomarkers with calibrated uncertainty, applied to the UK Biobank Pharma Proteomics Project.

  • Python
  • Survival analysis
  • Proteomics
  Repository Preprint

• sgVAMP

  The summary-statistics counterpart of gVAMP, extended to a multi-cohort setting for the joint estimation of shared and population-specific effects across ancestries.

  • Python
  • Summary statistics
  Repository

• gVAMPomi

  Approximate message passing for epigenome-wide association studies. Applied to the Generation Scotland methylation dataset it finds 92 trait-associated CpG probes conditional on all others, against 37 for the baseline MCMC approach.

  • Methylation
  • EWAS
  • Association testing
  Repository Abstract

• modular-Lan-MIA

  A modular library for benchmarking membership inference attacks against large language models, together with a document-level differential privacy auditing framework. Built during a machine learning and privacy internship at the University of Vienna.

  • Python
  • LLMs
  • Differential privacy
  Repository

• Single-cell LLM pipeline

  A Nextflow pipeline for fine-tuning and running inference with single-cell foundation models, built at a BioHackathon.

  • Nextflow
  • Python
  • Single cell
  Repository

Running gVAMP yourself? The gVAMP tutorial walks through compiling on an HPC cluster, inference under the Bayesian linear model, LOO/LOCO p-values, out-of-sample prediction, and running on DNAnexus via Docker.