mlchemad 1.1.0

Applicability domains for cheminformactics.

MLChemAD

Applicability domain definitions for cheminformatics modelling.

Getting Started

Install

pip install mlchemad

Example Usage

from mlchemad import TopKatApplicabilityDomain, data

# Create the applicability domain
app_domain = TopKatApplicabilityDomain()
# Fit it to the training set
app_domain.fit(data.training)

# Determine outliers from multiple samples (rows) ...
print(app_domain.contains(data.test))

# ... or a unique sample
sample = data.test[5] # Obtain the 5th row as a pandas.Series object 
print(app_domain.contains(sample))

Depending on the definition of the applicability domain, some samples of the training set might be outliers themselves.

Applicability domains

The applicability domain defined by MLChemAD as the following:

• Bounding Box
• PCA Bounding Box
• Convex Hull (does not scale well)
• TOPKAT's Optimum Prediction Space (recommended with molecular descriptors)
• Leverage
• Hotelling T²
• Distance to Centroids
• k-Nearest Neighbors (recommended with molecular fingerprints with the use of dist='rogerstanimoto')
• Isolation Forests
• Non-parametric Kernel Densities