SlitOR_data

Binary files and datasets for the SlitOR24-SlitOR25 paper:

Reverse chemical ecology in a moth: machine learning on odorant receptors identifies new behaviorally active agonists
Gabriela Caballero-Vidal [1], Cédric Bouysset [2], Jérémy Gévar [1], Hayat Mbouzid [1], Céline Nara [1], Julie Delaroche [1], Jérôme Golebiowski [2,3], Nicolas Montagné [1], Sébastien Fiorucci [2], & Emmanuelle Jacquin-Joly [1]
[1] INRAE, Sorbonne Université, CNRS, IRD, UPEC, Université de Paris, Institute of Ecology and Environmental Sciences of Paris, Versailles, France
[2] Université Côte d’Azur, CNRS, Institut de Chimie de Nice UMR7272, Nice, France
[3] Department of Brain and Cognitive Sciences, Daegu Gyeongbuk Institute of Science and Technology, Daegu 711-873, South Korea

Description

supp data 2 SlitOR24+25_dataset.xlsx

List of training and test set compounds with their associated name, CAS, SMILES, and activity.

descriptors_list_*.csv

CSV file containing the list of Dragon (v6.0.38) descriptors used by each model.

minmaxscaler_*.joblib

Saved MinMax scaler from the scikit-learn (v0.20.2) Python package (sklearn.preprocessing.MinMaxScaler).
To load the scaler, use the joblib Python package (joblib.load).

model_*.weka

Saved model from Weka (v3.8.2):

• weka.classifiers.lazy.IBk for OR25
• weka.classifiers.trees.RandomForest for OR24.

sphere-exclusion.py

Implementation of the sphere-exclusion algorithm. See the content of the file for the documentation.

Applicability Domain

For the similarity distance approach, the following constants (distance cutoff Dc and number of nearest-neighbors k) were used:

• OR24: Dc = 6.742, k=6
• OR25: Dc = 9.422, k=9

To obtain the reliability score for a molecule, given an array of descriptors, you can use the following code (implemented for Python v3.6.8 with numpy v1.16.1):

import numpy as np

def reliability_score(v, training_set, k, Dc):
    '''Reliability score for a molecule given its descriptors.
    Based on Tropsha A, Gramatica P, & Gombar VK (2003) QSAR & Combinatorial Science 22:69-77
    
    Parameters
    ----------
    v : np.ndarray
        An array of descriptors of shape (N_descriptors, ) for a molecule
    training_set : np.ndarray
        An array of descriptors of shape (N_molecules, N_descriptors) for the full training set
    k : int
        Number of nearest neighbors used to calculate the average distance
    Dc : float
        Distance cutoff
    
    Notes
    -----
    The arrays of descriptors should be normalized using the supplied MinMaxScaler
    '''
    # euclidean distances with training set
    distances = np.linalg.norm(v - training_set, axis=1)
    # sort distances
    distances.sort()
    # Average distance between the molecule and its k-NN in the training set
    D = distances[:k].sum() / k
    # reliability score
    return 1 + (D - Dc) / Dc