Here you can download in Silico models (QSAR and read across) for specific properties, or programs to build up your own model
With VEGA you can get a clear measurement of the reliability of the prediction, and also visualize the most similar compounds of your target substance, for read across assessment. VEGA has been used by ECHA to identify substances suspect to meet Annex III criteria for REACH.
QSARpy is an evolution of SARpy. It breaks the molecules of the training set and extracts modulators, fragments associated to a difference in the property/activity value. First, QSARpy verifies if the molecule to predict and one of the molecules of the training set share part of the structure, and one or more modulators can explain the differences between them. Then, QSARpy calculates the property/activity for the target adding the modulator(s) value to the value associate to the training set molecule.
The German Federal Ministry for the Environment, Nature Conservation, Building and Nuclear Safety recently funded a new project, called JANUS (FKZ 3716 65 4140), to prioritize and screen substances, considering PBT (persistent, bioaccumulative and toxic), CMR (carcinogenic, mutagenic and reprotoxic) sustances, as well as chemicals supposed to be endocrine disruptors. A consortium of three partners, Istituto di Ricerche Farmacologiche Mario Negri, Kode and KnowledgeMiner, will develop a platform within the JANUS project. JANUS further expands the previous prioritization project for PBT, called PROMETHEUS
SARpy breaks the chemical structures of the compounds in the training set into fragments of a desired size, and it identifies fragments related to the target property. It then also shows the fragments related to the effect. Inhibiting conditions are identified which prevent the appearance of the effect, even in presence of the active fragment. The system uses SMILES in the canonical form. It allows choice in building more conservative or more accurate models.
ToxDelta is a program which identifies the differences between two substances, and if these differences are related an increase or decrease of the effect. It nicely complements ToxRead. While ToxRead identifies the similar compounds and the common structural alerts between the target and similar compound, ToxDelta keeps into account the differences, and thus the dissimilarity features.
A preliminary BETA version is available for download.
CORAL breaks the chemical structures of the compounds in the training set into small components (SMILES attributes), based on the SMILES structure in the canonical form. The set of compounds are split into a subtraining set, a calibration set and a test set, in order to avoid overtraining. The inclusion of parameters such as the number of epochs, and rare SMILES attributes, enables the optimisation of the model.