PostHoc documentation

PostHoc is a Python toolkit for post-hoc variant attribution on GWAS-scale genotype data. It trains neural models on PLINK2 genotype matrices and produces PLINK-like output tables that can be compared or merged with standard GWAS pipelines.

PostHoc builds upon the neural-network attribution framework introduced by Yelmen et al. for identifying genome-wide association signals from artificial neural networks. In particular, PostHoc implements and extends the PAL (Post-hoc Attribution Loci) analysis described in that work within a modular, command-line framework designed for reproducible analysis of genotype data. See Concepts and methodology for details and citation.

The project currently includes:

• phenotype simulation on real genotype matrices

• baseline sparse logistic regression benchmarking

• Integrated Gradients SNP attribution

• PAL (Potentially Associated Loci) discovery with null-model significance testing

• PLINK2-compatible genotype input and GWAS-style outputs

• repeated-model analysis for robust locus discovery

Get started with Installation and Quickstart.

User guide

• Installation
  • Requirements
  • Install from source
  • Optional dependency sets
  • Verifying the install
  • GPU support
  • Next steps
• Quickstart
  • Prerequisites
  • 1. Simulate a phenotype with known causal SNPs
  • 2. Sanity-check with a linear baseline
  • 3. Train an MLP and compute SNP attributions
  • 4. Run PAL for robust, multi-model locus discovery
  • 5. Compare against ground truth
  • Next steps
• Input data formats
  • Genotype input (--pfile)
  • Phenotype input (--pheno)
  • Covariate input (--covar)
  • Sample alignment
  • Quality control filters
  • Output formats
• Output formats
  • simulate-pheno outputs
  • attribute output (.glm-style TSV)
  • pal output (TSV)
• Concepts and methodology
  • The overall approach
  • The MLP model
  • Integrated Gradients attribution
  • PAL and AMAS
  • Null model and significance testing
  • Practical implications for choosing options
  • Relationship to prior work

Reference

• Command-line reference
  • posthoc
  • simulate-pheno
  • baseline
  • attribute
  • pal
• Python API reference
  • posthoc.io
  • posthoc.qc
  • posthoc.models
  • posthoc.attribution
  • posthoc.simulation
  • Package overview

Project

• Development

Useful links

• Project repository

• Issue tracker

• README

License

PostHoc is distributed under the Apache License — see LICENSE in the repository.