A support vector machines approach for virtual screening of active compounds of single and multiple mechanisms from large libraries at an improved hit-rate and enrichment factor

Consensus model for identification of novel PI3K inhibitors in large chemical library

“…This is undesirable when performing virtual screening of large compound libraries [25,26]. Thus, this study has adopted the approach by Han et al [19] to generate putative inactive compounds to augment the negative training set. This method can generate putative negatives without requiring the knowledge of actual inactive compounds and studies had shown that classification models derived from these putative negatives can perform reasonably well in virtual screening [23,27].…”

“…To the best of our knowledge, this work is the first ligand-based virtual screening study for PI3K inhibitors. Studies have shown that models developed using a limited number of compounds are likely to have limited applicability domain [17,18], which may result in a large number of false positives when deployed for virtual screening of large chemical libraries [19]. Thus, a large number of compounds were used to develop the model so to expand the model's applicability domain.…”

“…Consequently, 65,377 compounds from 8,423 chemical families were used to develop a consensus model for the identification of PI3K inhibitors not specific to any isoforms. First, the true negative compounds were enriched with putative inactive compounds to increase the quantity and diversity of the negative set using the method developed by Han et al [19]. The significantly larger number of compounds in the training set will increase the applicability domain of the model and reduce the rate of false positives.…”

In-Silico Approaches to Multi-target Drug Discovery

“…We only evaluated dual-inhibitor search performance because of the availability of sufficient number of dual-inhibitors for conducting the tests and the relatively lower computational load for developing virtual screening models. SVMs were tested because of their good performance and high speed in screening large compound libraries (49) as well as our own experiences in developing SVM virtual screening tools (50,51). SVM for searching individual target and multi-target inhibitors is illustrated in Figs.…”

“…SVM of each individual kinase was developed by using 392∼1,303 known non-dual inhibitors published in the literature and 63,846∼66,214 putative non-inhibitors of EGFR, VEGFR, PDGFR, FGFR, Src and Lck respectively (representative compounds in PubChem and MDDR databases not known to inhibit each of these kinases respectively) by using the algorithm and procedure described in our earlier publications (50,51). The collective retrieval rate for each kinase pair was estimated by using 56∼188 known dualinhibitors of EGFR-FGFR, VEGFR-Lck, PDGFR-Src, and Src-Lck published in the literature, respectively.…”