Our methods
SUPERVISED LEARNING
Methods based on input information and the final output, and on searching for relationships between them (regression and classification methods such as linear and logistic regression, support vector machines, decision trees, Bayesian classifiers, multilayer neural networks, and image analysis).
UNSUPERVISED LEARNING
Methods that use only input information (without a final outcome/label) and look for relationships, similarities, and differences between the analyzed objects (e.g., clustering analysis, neural networks and self-organizing maps (SOM), principal component analysis (PCA), and expectation–maximization (EM) algorithms).
MOLECULAR DOCKING AND VIRTUAL SCREENING
It enables the selection of the most promising compounds for further in vitro and in vivo testing. The applied approaches support both structure-based drug design (SBDD) and ligand-based drug design (LBDD).
QSAR
ang. Quantitiative Structure-Activity Relationship
A method that quantitatively describes the relationship between chemical structure and properties: using numerical descriptors that encode information about a given chemical structure and its toxic effect, it is possible to build a mathematical model capable of predicting—in silico—the physicochemical properties, biological activity, and ADMET characteristics (absorption, distribution, metabolism, excretion, and toxicity) of compounds that have not yet been tested experimentally.
READ-ACROSS
Clustering and read-across approaches
A similarity-based approach: under the REACH guidance, a single group/category includes substances that are similar in their physicochemical, toxicological, and ecotoxicological properties, as well as in their chemical structure. Using these relationships, the properties of a target substance can be predicted by leveraging data available for one or more source substances. This makes it possible to generate knowledge about new, not-yet-tested substances based solely on their properties and chemical structure
WoE
ang. Weight of Evidence
The “weight-of-evidence” procedure refers to the process of collecting, weighing, and evaluating evidence in order to draw conclusions about a substance’s potential toxic effects. In this process, selected studies are screened against qualitative acceptability criteria, including their compliance with current OECD test guidelines and other standards, as well as their GLP status. A weight-of-evidence approach can be applied to assess virtually any toxicity endpoint, using data from both in vitro and in vivo studies.
