genoselect 0.3.0

scikit-learn-compatible genomic prediction and selection from SNP marker data

genoselect

scikit-learn-compatible genomic prediction and selection from SNP marker data.

Genomic selection is dominated by R packages (rrBLUP, BGLR, sommer). genoselect brings a clean, Pythonic, scikit-learn-compatible toolkit to the same problem: fit, cross-validate, and benchmark genomic-prediction models from 0/1/2 marker matrices, with a properly implemented GBLUP by REML at its core.

Features

• VanRaden genomic relationship matrix [vanraden_grm].
• GBLUP by REML — the Endelman (2011) spectral method, with equivalent ridge-regression marker effects for out-of-sample prediction (GBLUP).
• A common model interface — GBLUP, elastic net, random forest, gradient boosting, and a stacked super-learner ensemble, all as scikit-learn estimators (make_model, available_models).
• Breeding-relevant cross-validation — k-fold and leave-group-out, reporting predictive ability honestly (cross_validate, benchmark).
• GWAS — single-marker association (gwas) with PC structure correction, plus position-aware manhattan and qq_plot helpers.
• Read your own data — read_vcf, read_plink, read_hapmap load VCF / PLINK / HapMap files into 0/1/2 matrices (pure-Python, no extra deps).
• Simulation and QC — simulate_population, qc_markers, impute_markers.

Install

pip install genoselect

Or from source:

git clone https://github.com/mqfarooqi1/genoselect
cd genoselect && pip install -e .

Quick start

import genoselect as gs

# simulate a population with a heritable trait
pop = gs.simulate_population(n=250, m=800, n_qtl=40, h2=0.5, seed=1)

# fit GBLUP and inspect heritability + breeding values
fit = gs.GBLUP().fit(pop.geno, pop.pheno)
print(fit.h2_)            # estimated genomic heritability
print(fit.gebv_[:5])      # genomic estimated breeding values

# predict new genotypes
pred = fit.predict(pop.geno[:10])

# benchmark several models by cross-validation
cv = gs.benchmark(pop.geno, pop.pheno, k=5, random_state=1)
print(cv.summary())

Load your own data instead of simulating:

gd = gs.read_vcf("genotypes.vcf.gz")     # or read_plink("data"), read_hapmap("data.hmp.txt")
fit = gs.GBLUP().fit(gd.geno, phenotypes)

Because the estimators follow the scikit-learn API, they compose with the wider ecosystem (Pipeline, GridSearchCV, cross_val_score, …).

Methods

The genomic relationship matrix follows VanRaden (2008, doi:10.3168/jds.2007-0980); the mixed-model solver follows Endelman (2011, doi:10.3835/plantgenome2011.08.0024); the genomic-selection framework follows Meuwissen, Hayes & Goddard (2001, doi:10.1093/genetics/157.4.1819).

Tests

pip install -e ".[test]"
pytest

---

Author: Muhammad Farooqi · MIT licensed. A Python companion to the R package GSbench.