Interpretation of lazar predictions

Important: Do not trust these predictions blindly! Please evaluate each prediction carefully in light of your own toxicological knowledge. lazar is not a substitute for human experts, but a tool to assist human experts.

General principle

lazar derives its predictions from databases with experimentally determined toxicity data. To make a prediction for a new structure lazar searches in one of these databases for compounds with similar structures (neighbors) and calculates the prediction from their measured activities. The most important feature of lazar is, that chemical similarities are always determined in respect to a given toxic activity.

A detailled description and formal definition of the lazar algorithm has been published in:

• C. Helma: Lazy Structure-Activity Relationships (lazar) for the Prediction of Rodent Carcinogenicity and Salmonella Mutagenicity, Molecular Diversity 10, 147-158 (2006) [preprint]

The lazar source code is available from the Subversion repository svn+ssh://pdp8.in-silico.de/var/local/svn-pub/lazar/trunk (GNU GPL License)

The lazar webinterface tries to expose all information that has been used for a prediction, the exact meanings of the individual entries are described below.

lazar results

Prediction

The prediction for the selected compound and endpoint is displayed in the first row of the table with the following columns:

Predicted Activity (Confidence)

lazar prediction (active: red, inactive: green) with confidence index. Predictions are labelled as unreliable (yellow) if

• a compounds has very infrequent or unknown substructures
• the confidence index is lower than 0.025

Structure

2D representation of the query structure. Activating (red), deactivating (green) and unknown/infrequent (yellow) substructures are highlighted. NOTE: 2D rendering and substructure highlighting does not work correctly in all cases!

Measured Activity

If an identical structure has been found in the training set, predictions are made without any information from the query structure and the database activity is displayed here.

Additional Information

Relevant Fragments Display statistically significant and unknown/infrequent fragments for the current prediction.
Search Pubchem Retrieve additional information for the query structure from the PubChem database. PubChem will display an error message, if the current structure is not found!

SMILES/InChi

SMILES and InChi codes for the query structure. Useful for external database searches.

Similar Structures

A table with structurally similar compounds (neighbors) that have been used for the current prediction.

Similarity

Similarity in respect to the current endpoint , indicated by numbers between 1 (high similarity: white)) and 0 (no similarity: black).

Structure

2D representation of the neighbor. Activating (red), deactivating (green) and unknown/infrequent (yellow) substructures are highlighted. NOTE: 2D rendering and substructure highlighting does not work correctly in all cases!

Measured Activity

Experimental activity from the database.

Additional Information

Search Pubchem Retrieve additional information for the query structure from the PubChem database. PubChem will display an error message, if the current structure is not found!

SMILES/InChi

SMILES and InChi codes for the query structure. Useful for external database searches.