Knowledge Extraction

Knowledge Extraction (KE) is the process of discovering knowledge from data. KE is typically an iterative and interactive process which involves searching large volumes of data using a variety of techniques for valid and potentially useful patterns that can be considered knowledge about the data. In the field of drug discovery, the explosive growth of data generation methods over the past few years has lead to the generation of biological and chemical databases of increased size and complexity.

The knowledge discovered through KE can be used for two distinct purposes, interpretation and prediction. Interpretation provides explicit knowledge about phenomena described by potentially large, complex datasets, in a form that can be understood by a user. Prediction, or predictive modeling, focuses on forecasting events based on data describing events of the same type. In the case of prediction the interpretability of the knowledge extracted and used by the predictive models may be of secondary importance in which case the models are commonly known as "black box" models.

Noesis has implemented a collection of well-known and proprietary KE methods for both intepretation and prediction. The methods have been custom-designed for handling chemical and biological data in large volumes. Among the methods used are molecular clustering methods and substructure mining approaches useful for organizing and characterizing molecular data as well as classification and regression models for predictions of the biological behavior of novel, previously unseen molecules. Our software is designed to export the knowledge it produces in a standard form so that it can be preserved and shared within an interactive knowledge management system. Noesis also offers consulting services on KE system requirement gathering, specifications and design as well as customization services to extend its software to accomodate customers' individual needs.