Effect of doxycycline hyclate on corneal epithelial wound healing in the rabbit alkali-burn model. Preliminary observations.

We examined the effects of doxycycline hyclate on epithelial healing in vivo in the rabbit alkali-burn model. Twelve 2-3-kg Dutch belted rabbits were divided into three groups and received standard bilateral alkali burns (1 N sodium hydroxide for 30 s in an 11-mm circular plastic well). In group 1, two rabbits (four eyes) served as untreated controls. In group 2, five rabbits (10 eyes) received doxycycline hyclate (1.5 mg/kg) orally daily for 14 days. In group 3, five rabbits (10 eyes) received doxycycline hyclate (5 mg/kg) orally daily for 14 days. The epithelial defects were drawn and photographed on alternate days, after fluorescein staining. At conclusion, extracts of the corneas were evaluated for collagenase activity. At 14 days, the mean percentage of epithelial defects results in groups 1-3 were 50.0, 50.7, and 7.1%, respectively. Using the Wilcoxon rank sum test (two tailed), the differences were found to be statistically significant (p = 0.0015). Preliminary data indicated that oral doxycycline administration also decreased the collagenase activity in corneas obtained from these animals. Our preliminary findings indicated that systematically administered doxycycline hyclate, 5 mg/kg/day, promotes corneal reepithelialization in the rabbit alkali-burn model, a result, perhaps, of the drug's ability to inhibit excessive collagenase activity.

Exploratory data analytic techniques to evaluate anticancer agents screened in a cell culture panel.

Information theory is used to provide a measure of selectivity, i.e., the degree to which a drug has preferential toxicity or growth inhibition for one or a few cell lines from a large panel. The selectivity measure is intended to complement a measure of differential growth inhibition in evaluating the drug development potential of a new compound. Also, a similarity measure obtained from information theory is used to classify drugs according to their pattern of responses on the panel. Some structure-activity relations emerge. This work is applied to 176 agents selected to be tested by the National Cancer Institute in about 50 cell lines.

Clustering a large number of compounds. 2. Using the Connection Machine.

About 230,000 compounds in the National Cancer Institute Repository are available for screening under a new protocol. This paper is the second on an project to extract a representative sample of these compounds by clustering. The clustering program was implemented on the Connection Machine, a massively parallel computer with 16K processing elements. This implementation reduced a formidable task to a relatively routine run.

Clustering a large number of compounds. 3. The limits of classification.

Clustering is normally used to group items that are similar. In this application of obtaining a diverse sample from the 230,000 compounds in the National Cancer Institute Repository, we cluster to select compounds that are different from the rest, to optimize screening for new leads. With these constraints, our approach yielded many singleton clusters. We can interpret these results as evidence for a limit to classification, contrary to the customary view of chemistry as a study of classes of compounds.

Display and analysis of patterns of differential activity of drugs against human tumor cell lines: development of mean graph and COMPARE algorithm.

The objective of this study was to develop and investigate an approach to optimally detect, rank, display, and analyze patterns of differential growth inhibition among cultured cell lines. Such patterns of cellular responsiveness are produced by substances tested in vitro against disease-oriented panels of human tumor cell lines in a new anticancer screening model under development by the National Cancer Institute. In the first phase of the study, we developed a key methodological tool, the mean graph, which allowed the transformation of the numerical cell line response data into graphic patterns. These patterns were particularly expressive of differential cell growth inhibition and were conveniently amenable to further analyses by an algorithm we devised and implemented in the COMPARE computer program.

Clustering a large number of compounds. 1. Establishing the method on an initial sample.

The National Cancer Institute Division of Cancer Treatment has revised its drug-screening program. About 230,000 compounds in our repository are available for screening under the new protocol. This paper is the first on an attempt to extract a representative sample of these compounds by clustering. It reviews the establishment of the clustering method on a 4980-compound initial sample. The clustering algorithm is fairly simple. However, the molecular fragments employed to match the compounds are somewhat complex to distinguish a large number of compounds.

A two-component approach to predicting antitumor activity from chemical structure in large-scale screening.

A new way of combining physicochemical parameters such as the octanol/water partition coefficient (or log P) with molecular structure features has been devised to predict antitumor activity on a large, diverse set of compounds. This is done by adding the log P parameter as a separate component to an earlier method based on structure. The two-component approach is motivated by the dual concepts of accessibility and specificity. Extensive testing shows improvement in performance for the two-component method over the use of structure alone. All the compounds with definitive biological results in the in vivo NCI prescreen form the training set. The method separates the training set into disjoint subsets depending on the range of log P. Therefore, structure fragments receive activity weights that vary with the log P range. This change in weights accounts for the improved performance rather than any difference in the structural characteristics of the compounds in the different log P ranges.

Efficiency of antitumor screening relative to activity criteria.

Well over 100,000 compounds have been screened by the National Cancer Institute (NCI) in mouse leukemia P388 in the search for new leads toward antineoplastic agents. The protocol for screening requires confirmation testing of compounds showing initial activity. By activity criteria being varied for initial and confirmation testing, it is possible to improve the efficiency of screening. The experimental variability of P388 screening was first determined by analysis of outcomes from 90 tests on each of 21 compounds of varied activity. By application of these results to NCI P388 data for 31,428 compounds adequately tested on a daily times five schedule, the consequences of using different activity thresholds were estimated. The results here may apply to those of other similar large-scale screening programs.

Computer-aided selection of compounds for antitumor screening: validation of a statistical-heuristic method.

Recent work helped confirm the validity of a previously published statistical-heuristic method for selecting compounds to be processed in the National Cancer Institute (NCI) mouse tumor prescreen. One study involved about 35 000 compounds which satisfied certain biological and chemical criteria. These compounds, taken from the NCI collection, were ranked according to predicted probability of activity. The results showed 34% of the active compounds in the top 10% of the ranking. In a second study the predictive ability of a chemist and the computer were compared by having each rate almost 1000 previously unselected compounds for activity. The results were about equal with respect to the yield of biologically active compounds, though their selection agreement was fairly low.

A statistical-heuristic methods for automated selection of drugs for screening.

A statistical-heuristic method for selecting drugs for animal screening is developed with molecular structure features as predictors of biological activity. The method is intended to work on large amounts of data over varied structures. A trial of this method on a small data set allows some comparison with more sophisticated pattern recognition methods. Problems connected with interdependence among structure predictors are critical in this method and schemes to eliminate redundancy are reviewed. Alternate sets of structure predictors are considered. The discussion here outlines directions to be taken in the near future.

Selection of descriptors according to discrimination and redundancy. Application to chemical structure searching.

Descriptors, for our purposes, will be fragment screens in a chemical search system. Given a file of compounds, the discrimination of a set of descriptors can be defined in terms of their incidence and mutual incidence in the file. A theory is developed which provides both a heuristic for selecting descriptors and a method for evaluating their marginal discrimination. These ideas have been used to generate an efficient screen code for a large file of chemical structures.