EpistasisGaussianProcess¶

EpistasisGaussianProcess classifies genotypes as viable or nonviable with a Gaussian process classifier. Its draw over the other classifiers is calibrated probability: rather than a hard boundary, it returns a smoothly varying P(viable) with sensible uncertainty away from the training data. It uses the same add_gpm -> fit -> predict workflow as every other epistasis model.

In the figure each node is a genotype colored by P(viable) and sized by how certain the model is (|p - 0.5|); genotypes near the decision boundary stay small and pale, which is exactly the calibrated-uncertainty behavior a GP buys you.

When to use this model¶

Use EpistasisGaussianProcess when you want probability estimates you can trust as probabilities (for ranking, thresholding at a chosen risk level, or propagating uncertainty) rather than just a class label. It is the most expensive classifier here because fitting optimizes kernel hyperparameters, so prefer it on small to medium libraries.

How it works¶

1. An order-1 EpistasisLinearRegression (model.additive) learns each mutation's additive contribution from the continuous phenotypes.
2. The design-matrix columns are scaled by those additive coefficients.
3. sklearn's GaussianProcessClassifier (RBF kernel by default) is fit on the projected matrix with binarized labels (y > threshold -> 1).

Constructor parameters¶

threshold (float, required)

Phenotype cut-off. Genotypes above it are class 1 (viable).

model_type (str, default "global")

Design-matrix encoding: "global" (Hadamard) or "local" (biochemical).

kernel (sklearn.gaussian_process.kernels.Kernel | None, default None)

Covariance kernel. Defaults to an RBF() kernel. Pass your own to encode different smoothness assumptions.

n_restarts_optimizer (int, default 0)

Number of restarts for the kernel-hyperparameter optimizer. Raise it for a more thorough (slower) fit.

max_iter_predict (int, default 100)

Iterations for the Laplace approximation used at prediction time.

random_state (int | None, default None)

Seed for reproducible optimization.

Workflow¶

import numpy as np
from epistasis.models.classifiers import EpistasisGaussianProcess

model = EpistasisGaussianProcess(
    threshold=float(np.median(gpm.phenotypes)),
    random_state=0,
)
model.add_gpm(gpm)
model.fit()

p_viable = model.predict_proba()[:, 1]  # calibrated P(viable)
labels = model.predict()                # 0 / 1
accuracy = model.score()

Key methods¶

See also EpistasisLogisticRegression, discriminant analysis, and EpistasisGaussianMixture.