Fuzzy Logic Interval Clustering for Drug Discovery PREDICTION - PowerPoint PPT Presentation

Fuzzy Logic Interval Clustering for Drug Discovery

PREDICTION ACCURACY FOR DRUG DISCOVERY • PROBLEM Hundreds of Test EACH thousands of compoundon – High throughput screening compoundsare the desired (HTS) is expensive and time- available protein consuming – Artificial Neural Network (ANN) can detect patterns, Does the but prediction accuracy is low compound bind with the protein • SOLUTION Yes No – KU invention integrates Fuzzy Logic and Interval Clustering (FLIC) to the Active Non-Active existing ANN operation – Higher prediction accuracy of Active Compounds = cheaper, faster process

HOW IT WORKS • ANNs simulate the structure and function of a brain using interconnected nodes – Trained using known (labeled) data • Fuzzy Logic applies confidence levels, rather than a binary true/false classification • Interval Clustering groups items based on “membership values” • KU innovation: FLIC is used to train and validate many ANNs and the best performing ones are Basic ANN Structure chosen to increase efficiency and accuracy – Clustering is used during training – The most accurate ANNs are chosen during validation to process the unknown data – During testing, each ANN calculates a probability value that indicates how “active” or how “inactive” each compound is • The higher the number of ANNs that identify a compound as “active”, the higher the chances it is actually active. Clustering

FEATURES AND BENEFITS • FEATURES – FLIC utilizes “fuzzy logic and interval clustering” • Improves overall performance and efficiency in predicting active compounds • Identifies existence of misclassification in training data – Easily integrated into available drug discovery process – Patent pending (2015) – Proof of concept completed using real datasets from KU HTS lab • 50x improvement over ANNs alone • BENEFITS – Identifies active compounds at a low number of misclassifications – Moves the process from the wet lab to a dry lab – Reduces cost and provides faster results • Method could be applied to other types of large datasets for pattern recognition (facial or speech recognition, medical diagnosis)

CONTACT INFORMATION David Richart, JD Licensing Associate KU Center for Technology Commercialization University of Kansas drichart@ku.edu (785) 864-0124 Read More: http://kuic.ku.edu/available-technologies/high-throughput-screening-HTS-accurate-lower-cost-faster-results