Ensemble modeling

Numerous QSAR tools have been developed [152, 154] and used in modeling physicochemical data. These vary from simple linear to more complex nonlinear models, as well as classification models. A popular approach more recently became the construction of consensus or ensemble models ( combinatorial QSAR ) combining the predictions of several individual approaches [155]. Or, alternatively, models can be built by rurming the same approach, such as a neural network of a decision tree, many times and combining the output into a single prediction. 

Bradley and coworkers used the 3D pharmacophore ensemble model to filter a virtual combinatorial library of 3924 N-substituted glycine peptoids (30) containing three known a, actives down to a set of 639 products. Using a cut-down technique, a 160 compound combinatorial library was designed in which the number of compounds that passed the ensemble model filter was maximized. This library contained two of the three known actives present in the original 3924 compound virtual library. This represents a substantial enrichment [(2 actives/160 products) X 100 = 1.25% vs (3 actives/3924 products) x 100 = 0.076%]. 

An attempt to build QSPR models for large structurally diverse sets of crown ethers and to extract the information hidden in the experimental data about the mac-rocyclic effect (ME) was made by Varnek and Solov ev et al.25,77 using an ensemble modeling approach and SMF descriptors. The calculations were performed on... 

By combining elements of the conformational ensemble model for protein structures in solution (figure 7.6) with the observations that flexibility-related differences in /ceat occur... 

At this level of description, this ensemble model assumes homogeneous doping and excitation densities such that the entire ensemble of ions can be described as having identical environments at all times, and the entire ensemble participates equally in the photodynamics of the system. 

These techniques span the entire field from multiple linear regression (MLR)-type methods and various forms of neural network architectures to rule-based techniques of different kinds. These approaches also span from single models to multiple models, that is, consensus or ensemble modeling. Terms like machine learning and data or information fusion are also frequently encountered in this area of research, as well as the concepts of applicability domain and validation. 

Various forms of these combinations with an accuracy of the ensemble models better than the corresponding single reference have been reported [90, 114—119]. 

When nsing ensemble modeling, that is, calculating multiple models with the same techiuque, each model might use a different subset of descriptors. An alternative is to do the descriptor subset selection first. ... 

J.D. Woods, R. Onken (1982). Diurnal variation and primary production in the ocean - preliminary results of a Lagrangian ensemble model. J. Plankton Res., 4, 735-756. 

Set-Wise Similarity reuse much of the machinery developed to calculate background Ensembles models and extend it to calculate E-values. By exhaustively... 

Norman, A.W., Mizwicki, M.T. and Norman, D.P. (2004) Steroid-hormone rapid actions, membrane receptors and a conformational ensemble model. Nature Reviews Drug Discovery, 3,... 

U. Norinder, P. Liden, H. Bostrom, Prediction of aqueous solubility using rule-based systems (RDS, www.compumine.com) and ensemble modelling, unpublished results. 

Figure 4 Monte carlo and ensemble model predictions of the dependence of the rate of ethene hydrogenation on particle size. Data are presented for T=280K, hydrogen pressure = 100 Torr, ethene pressure = 10 Torr...

Figure 10.16 Plot of average binding energy as a function of cluster size for the (NH3) H + system. The open squares are literature values, the circles are from the Engelking equation, while the triangles are from the Klots vaporization ensembles model. The data were normalized at n = 7. Adapted from Wei et al. (1990).

The evaporative ensemble model is one in which the cluster ions have been generated in a broad distribution of cluster sizes and internal energies. Hence, the rate constant, k E), is inversely related to the reaction time, f, so that ln( ) = —ln(t) and cnn(A ) = —dln(t). This permits us to convert the derivative in Eq. (10.55) from dln(fe) to dln(f) (Lifshitz, 1993) as... 

The induced-fit model shown in Figure 7.10(b) allows for some flexibility, but still focuses in on a single unbound receptor conformation. In reality, proteins are in constant motion between different conformational states with similar energies as stated in an outstanding review of protein flexibility in drug design." So the best model these days is the conformational ensemble model shown in Figure 7.10(c). Here a number of protein conformers exist in... 

---