Five test tubes of diluted organism for each isolate

und in the MD’s full trajectory (Table 3). For instance, Fig 4a and 4b show Cavity Attributes and Protein RMSD data sets having better SQD values than Cavity RMSD data set for both partitioning methods. However, Table 3 indicates the high difference on the average and variance between the best partitions from these methods and the MD’s full trajectory, which in turn reduces favorable representativeness to the MD trajectory. These inaccurate values can be explained by the fact that partitioning methods work well for finding spherical-shaped clusters in small to medium-sized dataset [43]. The practical consequence of this is represented by the graphs from Fig 4, where the lines come up and down and show slight differences in the SQD values between the data sets as the number of clusters varies. Unlike partition-based clustering methods, hierarchical algorithms appear to outperform the partitioning for all data sets and present more stable SQD values after a certain number of clusters (Fig 5). As stated earlier, small numbers of medoids are unable to Celgosivir PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/19667359 provide large width and are therefore not able to achieve good level of similarity to the quartile values of the MD’s full trajectory. This is evidenced by the high SQD values in the beginning of the graphs (a), (b) and (c) from Fig 5. The Ward’s method presents low SQD values in partitions with PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/19667298 16, 22 and 25 clusters for the Protein RMSD data set. The average and variance values of medoids from these partitions are: -6.62 and -0.51 for k = 16; -6.62 and -0.52 for k = 22; and -6.61 and -0.53 for k = 25. Similar to partitioning methods, these statistical values are far away from the same statistical values found in the MD trajectory and we may conclude that such solutions are un