QSAR in drug discovery
Alexander Tropsha · Cambridge University Press eBooks · 2010
INTRODUCTION With nearly fifty years of rich history of methodology developments and applications (the Hansch article of 1963 is often considered first in the field), quantitative structure/activity relationship (QSAR) modeling is a well-established area of research. As is true perhaps for any computational field, QSAR modeling has been both blessed and sometimes cursed in the literature. In the first volume of the famous book series titled Reviews in Computational Chemistry , Boyd summarized several documented cases when QSAR modeling was instrumental in discovering new drugs of drug candidates in advanced phases of clinical trials. The methodologies used by that time were relatively simple, employing a small number of physical chemical descriptors and statistical methods such as multiple linear regression. QSAR modeling was viewed solely as a tool for lead optimization; that is, it was employed to elucidate the relationship between structure and activity in relatively small congeneric compound series and predict relatively small structural modifications leading to enhanced activity. Since the late 1980s the field has changed dramatically, fueled by changes in the size, complexity, and availability of experimental data sets of biologically active compounds. These changes have been coincidental with the advances in chemometrics, resulting in a significant increase in the number of chemical descriptors as well as growing implementation of machine learning and advanced statistical modeling techniques available for QSAR studies.