lazar

Ruby libraries for the lazar framework

Dependencies

lazar depends on a couple of external programs and libraries. All required libraries will be installed with the gem install lazar command. If any of the dependencies fails to install, please check if all required development packages are installed from your operating systems package manager (e.g. apt, rpm, pacman, …). You will need a working Java runtime to use descriptor calculation algorithms from CDK and JOELib libraries.

Installation

gem install lazar

Please be patient, the compilation of external libraries can be very time consuming. If installation fails you can try to install manually:

git clone https://github.com/opentox/lazar.git cd lazar ruby ext/lazar/extconf.rb bundle install

The output should give you more verbose information that can help in debugging (e.g. to identify missing libraries).

Tutorial

Execute the following commands either from an interactive Ruby shell or a Ruby script:

Import lazar

require 'lazar' include OpenTox

Create and use lazar models for small molecules

Create and validate public lazar models

public_models = Import.public_data

This command creates models for all training data in the data folder and validates them with 5 independent crossvalidations. This may take some time (several hours). Retrieve validation results with public_models.crossvalidations.

Make predictions

Single compounds

Create a compound

compound = Compound.from_smiles "NC(=O)OCCC"

Select a model

model = public_models.first

model.predict compound

Batch predictions

Create a CSV file with one or two columns: An optional Substance ID and SMILES codes for the substances to be predicted. The first line should contain either “ID,SMILES” or just “SMILES” if there is no ID column.

Import the dataset

dataset = Dataset.from_csv_file batch_file.csv

Select a model

model = public_models.first

Make a batch prediction

prediction_dataset model.predict dataset

View predictions

prediction_dataset.predictions

Create and validate models from your own datasets

Create a training dataset

Create a CSV file with two or three columns: An optional Substance ID, SMILES and toxic activities (qualitative or quantitative). Use -log10 transformed values for quantitative values. The first line should contain “ID” (optional), SMILES and the endpoint name. Add metadata to a JSON file with the same basename containing the fields “species”, “endpoint”, “source”, “qmrf” (optional) and “unit” (regression only). You can find example training data in the data folder of lazar.

Create and validate a lazar model with default algorithms and parameters

validated_model = Model::Validation.create_from_csv_file training_data.csv

This command will create a lazar model and validate it with five independent 10-fold crossvalidations. You can use the model in the same way as the public models.

Create and validate models from PubChem Assay Data

If you know the PubChem Assay ID (AID), you can create and validate models directly from PubChem.

Download datasets from PubChem

csv_file = Download.pubchem_classification aid: 1205, species: "Rodents", endpoint: "Carcinogenicity", qmrf: {group: "QMRF 4.12. Carcinogenicity", name: "OECD 451 Carcinogenicity Studies"}

or

csv_file = Download.pubchem_regression aid: 1195, species: "Human", endpoint: "Maximum Recommended Daily Dose", qmrf: {group: "QMRF 4.14. Repeated dose toxicity", name: "OECD 452 Chronic Toxicity Studies"}

This will create new CSV and metadata files in the data folder (or update existing ones). Regression data will use -log10 transformed molar values. Use this file either with Model::Validation.create_from_csv_file or create all models in the public folder with Import.public_models.

Experiment with other algorithms

You can pass algorithm specifications as parameters to the Model::Validation.create_from_csv_file and Model::Lazar.create commands. Algorithms for descriptors, similarity calculations, feature_selection and local models are specified in the algorithm parameter. Unspecified algorithms and parameters are substituted by default values. The example below selects

  • MP2D fingerprint descriptors
  • Tanimoto similarity with a threshold of 0.1
  • no feature selection
  • weighted majority vote predictions

``` algorithms = { :descriptors => { # descriptor algorithm :method => “fingerprint”, # fingerprint descriptors :type => “MP2D” # fingerprint type, e.g. FP4, MACCS }, :similarity => { # similarity algorithm :method => “Algorithm::Similarity.tanimoto”, :min => [0.5,0.2] # similarity thresholds for neighbors: first value for predictions with high confidence, second value for predictions with medium confidence }, :feature_selection => nil, # no feature selection :prediction => { # local modelling algorithm :method => “Algorithm::Classification.weighted_majority_vote”, }, }

training_dataset = Dataset.from_csv_file “hamster_carcinogenicity.csv” model = Model::Lazar.create training_dataset: training_dataset, algorithms: algorithms ```

The next example creates a regression model with

  • calculated descriptors from OpenBabel libraries
  • weighted cosine similarity and a threshold of 0.5
  • descriptors that are correlated with the endpoint
  • local partial least squares models from the R caret package

algorithms = { :descriptors => { # descriptor algorithm :method => "calculate_properties", :features => PhysChem.openbabel_descriptors, }, :similarity => { # similarity algorithm :method => "Algorithm::Similarity.weighted_cosine", :min => [0.5,0.2] }, :feature_selection => { # feature selection algorithm :method => "Algorithm::FeatureSelection.correlation_filter", }, :prediction => { # local modelling algorithm :method => "Algorithm::Caret.pls", }, } training_dataset = Dataset.from_csv_file "EPAFHM_log10.csv" model = Model::Lazar.create(training_dataset:training_dataset, algorithms:algorithms) Please consult the API documentation and source code for up to date information about implemented algorithms:

You can find more working examples in the lazar tests.

Create and use lazar nanoparticle models

eNanoMapper import is currently broken, because API and data models change unpredictably and we have no resources to track these changes. Please contact [email protected], if you want to fund the further development of nanoparticle models

Create and validate a nano-lazar model from eNanoMapper with default algorithms and parameters

validated_model = Model::Validation.create_from_enanomapper

This command will mirror the eNanoMapper database in the local database, create a nano-lazar model and validate it with five independent 10-fold crossvalidations.

Inspect crossvalidation results

validated_model.crossvalidations

Predict nanoparticle toxicities

Choose a random nanoparticle from the “Potein Corona” dataset training_dataset = Dataset.where(:name => "Protein Corona Fingerprinting Predicts the Cellular Interaction of Gold and Silver Nanoparticles").first nanoparticle = training_dataset.substances.shuffle.first

Predict the “Net Cell Association” endpoint

validated_model.predict nanoparticle

Experiment with other datasets, endpoints and algorithms

You can pass training_dataset, prediction_feature and algorithms parameters to the Model::Validation.create_from_enanomapper command. Procedure and options are the same as for compounds. The following commands create and validate a nano-lazar model with

  • measured P-CHEM properties as descriptors
  • descriptors selected with correlation filter
  • weighted cosine similarity with a threshold of 0.5
  • Caret random forests

algorithms = { :descriptors => { :method => "properties", :categories => ["P-CHEM"], }, :similarity => { :method => "Algorithm::Similarity.weighted_cosine", :min => [0.5,0.2] }, :feature_selection => { :method => "Algorithm::FeatureSelection.correlation_filter", }, :prediction => { :method => "Algorithm::Caret.rf", }, } validation_model = Model::Validation.from_enanomapper algorithms: algorithms

Detailed documentation and validation results for nanoparticle models can be found in this publication.

Documentation

Copyright (c) 2009-2018 Christoph Helma, Martin Guetlein, Micha Rautenberg, Andreas Maunz, David Vorgrimmler, Denis Gebele. See LICENSE for details.