Knowledge Discovery in High Throughput Screening: Towards Creation of a Data Mining Techniques Selection Guideline

Abstract

High Throughput Screening (HTS) is a well-established process in which a large amount of potential effectors of certain biological activity are tested in assays using automated instruments. Nowadays it has found its place in both drug development within pharmaceutical industry and fundamental biological research including functional genomics and cell biology. It provides an effective tool to identify a very small number of active effectors, normally 5%, from a large number of inactive ones. However, datasets generated by each HTS experiment usually contain hundreds of thousands of records with a high dimensionality, especially when it is designed to monitor cellular changes. Therefore, it has come to biomedical researchers' attention that advanced computational methods, such as data mining techniques, are required in order to make full use of the collected HTS datasets. In addition, because the number of available data mining techniques is quite large and each data mining technique owns unique characteristics, the appropriate use of data mining techniques to analyze big HTS data has become one of the big concerns for HTS data analysts. The objective of this master graduation project is to develop a DMTs selection framework which guides analysts to the right DMTs during the HTS knowledge discovery process. The end-product of this research is called HTS-DMTs Framework which transforms abstract DMTs selection knowledge into a practical decision supporting framework. The HTS-DMTs Framework is visualized as two tables. The first table offers a complete overview of the HTS data mining tasks and provides a list of applicable DMTs for each of these data mining tasks. In the second table, characteristics and supporting software of all the DMTs in the first table are presented and such practical information could help HTS data analysts specify the most appropriate DMTs for a HTS data mining task. The developed framework is evaluated by a case study. In the case study two HTS datasets from the Cell Screening Center at UMC Utrecht are collected and analyzed with the guidance of the framework. Although the research follows the guidelines of case study defined by Yin (2008), the process is modified according to the quantitative characteristic of this research. Specifically, the case study puts more emphasis on showing the process of HTS data analysis using DMTs. This evaluation has demonstrated effectiveness of the framework. It helps analyst to pinpoint the most appropriate DMTs during a HTS analysis process.