Benchmark studies and further validation of API technology have been reported recently (See Abstract). These studies included construction of a large structural database (N=1470) of candidate estrogens tested in laboratory animals by the National Institutes Of Health (NIH). The NIH study which was published in 1968 is the most comprehensive study available to date, took approximately ten years to complete and required over 100,000 animals. More than 100 pharmaceutical companies and academic centers provided the compounds in the study. Using API's estrogen search engine, candidate estrogens were identified in the database within a few days. All compounds determined to be active in the NIH study were identified in API's search. Compounds which were prodrugs did not hit the search engine but the known active metabolites were hits. The overall enrichment of active compounds found by the search engine was between 30 and 60 fold (see below). PDC-7 which was prospectively designed, synthesized and shown in animal studies to be a highly potent estrogen also hit the search engine. In addition to successfully separating active from inactive compounds, the search engine was able to select compounds with the highest estrogenic activity (see below).

These results validate the accuracy, precision and power of API's search engine technology as well demonstrate the potential cost savings of this approach. Namely, if conducted today, the NIH study would be estimated to cost hundreds of millions of dollars and an almost prohibitive amount of time. By identifying candidates before synthesis and animal testing, API's in silico technology would have saved years of experimental studies and obviated greater than 95% of the cost of the project.

Similar to the benchmark study of the estrogen search engine described above, a second study has been completed using the androgen search engine with compounds tested in vivo for androgenic activity. The androgen biological testing results were published in 1964 by NIH and are a comprehensive compilation of animal testing over seven years encompassing approximately 1000 chemicals derived from multiple sources including pharmaceutical companies and academic institutions. API constructed a three dimensional structural database of the compounds and their analogs (N=1345) and conducted a search of the database using the androgen search engine. All compounds found by NIH to have androgenic activity hit the search engine. Compounds which were prodrugs did not hit the search engine but the known active metabolites were hits. The overall enrichment of active compounds found by the search engine was above 30 fold. In addition to successfully separating active from inactive compounds, the androgen search engine was able to select compounds with the highest androgenic activities (see Figure below). Compounds with estrogenic activity did not hit the androgen search engine nor did compounds with androgenic activity hit the estrogen search engine. These results demonstrate the specificity of the search engines as further delineated below. Taken as a whole, the benchmark studies of the estrogen and androgen search engines demonstrate the power of APIâ??s computational methods to rapidly and accurately assess biological activity of a given compound or large database of compounds. In the case of the NIH studies, APIâ??s proprietary search engines provided information within a few hours compared with the decades of research and development devoted to these drugs.

Cross-validation studies of the estrogen, androgen, antidepressant, sedative and antibiotic (Anthrax) search engines were recently published (See Abstract). Each of the search engines was used to search a combined database containing compounds with widely varying structures and a spectrum of known biological activities. The results of the search engines demonstrate considerable specificity in that the search engines correctly identified compounds with known biological activities and did not identify compounds that did not have such activities. Compounds structurally unrelated to the compounds (standards) used to create a given search engine were also correctly identified to have a given known activity. Thus, the search engine hits are not limited to compounds that are only structurally related to the standards. In cases where multiple biological activities were known for a given compound, several search engines correctly identified the structure. In some cases, the search engines identified compounds not known to be tested for a given activity. While biological activity is not assured for these compounds, such data provide information as to which drug candidates should be tested further for a given activity (e.g. a potential side effect).

The search engines were derived from the fit of active ligands into a scaffold derived from partially unwound gene sequences in double stranded DNA and/or gene/nuclear protein interfaces. Sybyl/Unity molecular modeling software (Tripos Associates, Inc, St. Louis, MO) was employed. Each search engine is composed of: 1) electrostatic spatials which reflect positions in space of hydrogen bonds between the ligands and macromolecules; 2) an excluded volume which represents the macromolecule surface; and 3) an included volume that is the composite surface of active ligands aligned by fit into the site. In order to be considered a hit, a candidate molecule must contain hydrogen bonding functional groups that fit within the electrostatic spatials and fit within the included volume without violating the excluded volume. Please note this section is currently under construction and is being updated and expanded.