The present invention employs a novel use of logic and data analysis tools to improve the efficiency of drug discovery.
High Throughput Screening (HTS) is an automated process of testing a large number of compounds in a wet lab setting to compose a list of active compounds. Prediction of likely active compounds reduces the number of compounds that need to be tested, which reduces the time and cost of such testing. Artificial Neural Network (ANN) models are currently used to achieve this, but their prediction accuracy is low. The present invention, when used in conjunction with current methods, significantly increases their accuracy.
Software employing the methods of this invention can be easily incorporated into existing HTS methods that utilize ANN prediction models.
ANN models require a training phase to identify patterns in a set of tested compounds. New compounds are then compared to these patterns to predict whether a particular compound is likely to be active or non-active. The present invention uses a mathematical algorithm to improve this training by creating clusters based on fuzzy logic. This adds another dimension to the ANN methods by assigning a confidence level to each compound.
The primary benefit of this invention is a significant improvement in the prediction accuracy of current drug discovery screening methods. This directly leads to improved screening efficiency and reduced drug discovery costs.
Compared to traditional HTS methods that use only ANN or other machine learning techniques, this invention can provide results that are 50 times more accurate.
Though the software was designed for use in drug classification, the algorithm at its core could be used in other pattern classification scenarios. The inventors have contemplated its use in automated speech recognition, face recognition, and medical diagnosis applications.