ActiPred

ActiPred is a tool for the automatic prediction of properties of graphs. The user can provide the tool with training graphs that share a common property. The tool then builds a model based on this training data and can subsequently predict the property for any given graph with high accuracy.

ActiPred/Chem       Because the "Why" is Important

In the chemical space, ActiPred/Chem can be used to predict the activity of compounds. It has been tested on a variety of activity classes including toxicity, blood-brain barrier permeability, mitochondria respiratory electron transport system inhibition, human intestine absorption, and more. Since ActiPred works directly in the graph space, its classification results can be significantly better than traditional fingerprint-based classifiers. Some sample results can be found here: → ActiPred Quality

Besides competitive prediction performance, ActiPred provides the unique ability to indicate "why" a positively classified compound is active. For this purpose, ActiPred matches fragments of the classified compound with the most relevant actives from a prediction-specific database of actives and inactives. Subsequently, it highlights the most relevant matches of the classified compound since these will be the ones that contibute to the activity with the highest likelihood.

An illustrative example for the ActiPred/Chem input and output is shown below.

Besides providing a classification and a score, ActiPred/Chem also highlights substructures of each active molecule to indicate which part of the molecule contributed with highest likelihood to its activity. Similarly, ActiPred highlights for each inactive molecule the parts that contributed with highest likelihood to its inactivity, using the same logic as described earlier.

Currently, ActiPred/Chem ships with the following fully integrated and tested prediction models (all support highlighting):

ActiPred/Chem can be easily integrated into existing workflows as it supports both MOL2 and SDF data formats.