In silico methods for toxicity prediction

Author: Christoph Helma
Affiliation:in silico toxicology gmbh
Date: 2012-10-04


In silico methods

Systems biology/molecular modeling

Model individual events (e.g. receptor interactions, (de)toxification) of the adverse outcome pathway

Examples: VirtualToxLab/Biograf

Expert systems

Formalize expert knowledge about chemicals and toxicity mechanisms and create a software program

Examples: Derek/Lhasa

Data driven

Use all existing data for a particular endpoint and apply machine learning/QSAR algorithms in order to create a prediction model

Examples: Classical QSARs, Topkat, Multicase, lazar

Lazy-Structure-Activity Relationships (lazar)

Automated read across predictions

Lazar estimates the confidence (applicability domain) for each prediction

Chemical Similarity

Can be based on

Lazar uses activity specific similarities

Activity specific similarities

Consider only relevant (i.e. statistically significant) substructures, properties, ... for similarity calculations

Algorithms for finding relevant substructures (by A. Maunz):





Lazar limitations

in silico toxicology gmbh

Open source software and algorithm development

Why open source?

EU Research projects (FP6/7)

Sens-it-iv:Novel testing strategies for in vitro assessment of allergens
Scarlet:Network on in silico methods for carcinogenicity and mutagenicity
OpenTox:Open source framework for predictive toxicology
ToxBank:Integrated data analysis and servicing of alternative testing methods in toxicology
ModNanoTox:Modelling toxicity behaviour of engineered nanoparticles

Free products and services

Lazar application:
OpenTox Webservices:
Source code:

Issue tracker, documentation, ...

Commercial products and services

Commercial products and services