The team behind TOXsIgN has invested considerable time, resources, and effort into collecting, curating, processing, developing, and maintaining this web resource. Please follow the citation guidelines if you are using it.
Signature prediction tool
This tool allows users to predict the properties of a chemical by feeding its signature to a neural network trained on the ChemPSy dataset.
Predictions include the probability that the chemical belong to a chemical clusters.
How to perform the prediction Input data: Select the model you wish to use and the signature using the "Select signature" selector.
Output data: Model-dependent table.
Prediction model performance
Prediction models have several performance metrics available, defined during the training:
Cluster-based models will try to predict whether an experimental condition is in a cluster or not
Precision is the measure of How many conditions the model labelled as part of a cluster are actually part of it? It is the ratio of True positives/(True positives + False positives)
Recall is the measure of Out of all the conditions in a cluster, how many did the model correctly label? It is the ratio of True positives/(True positives + False negatives)
Specificity is the measure of Out of all the conditions not in a cluster, how many did the model correctly label? It is the ratio of True negatives/(True negatives + False positives)
The "best" metrics depends of the user needs
Probability is not a metric of the model, but simply the probability that a condition belongs to a cluster