DTP Branches and Offices
Introduction
Prediction of Biochemical Mechanism of Action from the In Vitro Antitumor Screen of the National Cancer Institute
Kenneth D. Paull, Ernest Hamel, and Louis Malspeis
We are in an age when experimental data can be analyzed with unprecedented ease and speed. This permits entirely new approaches to large accumulations of data. The new National Cancer Institute (NCI) in vitro anticancer screening program generates prodigious quantities of information-laden biological test results, captured in an easily accessible computerized database. We describe an analysis of the NCI in vitro database that answers questions that could not be asked in any previous screening program.
The routine implementation of the current NCI in vitro anticancer screen to evaluate the efficacy of synthetic compounds and natural products was initiated in April 1990. The screen, developed over a period of several years (1, 2, 3) by the Developmental Therapeutics Program (DTP), employs 60 human tumor cell lines that have been grouped in disease subpanels including leukemia, non-small-cell lung, small-cell lung, central nervous system, colon, melanoma, ovarian, and renal tumors or cell lines. In December 1993, several changes were made to this panel of cell lines. A new panel of eight breast lines and a new panel of two prostate lines were added. To keep the panel total at 60, 10 lines identified as redundant or technically difficult to use (e.g., the two small cell lung lines) were dropped. Approximately 1000 compounds and natural product extracts have been screened each month. The decision that there would be no alteration of the assay method and cell lines employed in the screen for more than 12 months has permitted the evaluation of thousands of compounds under identical conditions.
Our analysis of these data is performed by a program we call COMPARE (2, 3, 4, 5, 6, 7, 8). A probe or “seed” compound can be specified by using the compound’s NCI accession number (the NSC number). The COMPARE algorithm then proceeds to rank an entire database in the order of the similarity of the responses of the 60 cell lines to the compounds in the database to the responses of the cell lines to the seed compound. Similarity of pattern to that of the seed is expressed quantitatively as a Pearson correlation coefficient (PCC). The results obtained with the COMPARE algorithm indicate that compounds high in this ranking may possess a mechanism of action similar to that of the seed compound (6, 7).