Diversity analysis of 14 156 molecules tested by the National Cancer Institute for anti-HIV activity using the quantitative structure-activity relational expert system MCASE.

Using the MCASE program, a procedure to analyze the diversity of the large amount of available HIV-1 antiviral data was proposed. A subset of 1 819 chemicals was logically selected from the original 14 156 chemicals tested by NCI. This subset of chemicals was shown to contain most of the structural and the functional information of the original database. A full analysis of the 1 819 chemicals by the MCASE program produced a correlation between chemical structures and HIV antiviral activity. In our model, 74 fragments were identified as being responsible for all the chemical's HIV antiviral activity. These fragments may be related to different inhibiting mechanisms, some known and some probably still unknown. The expert system resulting from this analysis can be used to predict the activity of new chemicals and to design new agents that can target multiple enzymes. This was shown to be the case by using the model to predict the activity of 10 diverse chemicals whose activities were not known at the time of model development. Of these, 8 were predicted in agreement with experimental observations. As far as we can tell, this is probably the first project ever to attempt to create a quantitative model of activity for such a massive database of diverse chemicals.