Biomedical Engineering Reference
In-Depth Information
been put on combining different clustering approaches or integrating them with other
computational methods. For example, clustering has been combined with principal
component analysis to add structural diversity criteria to compound selection [72].
Furthermore, cluster analysis has recently been integrated with probabilistic models
of activity to incorporate biological activity information in the clustering process
[73]. Moreover, clustering has been combined with molecular network analysis to
classify active compounds [74]. This approach is also applicable to LBVS.
15.8 MACHINE LEARNING
Machine learning approaches for supervised learning have become increasingly popu-
lar for LBVS. In general, learning sets of known active compounds (positive training
examples) and inactive compounds (negative training examples) are used to build
models of activity for class label prediction. Confirmed inactive compounds might
be taken from HTS data, but randomly selected database compounds assumed to
be inactive are often used as negative training examples. For LBVS, the most pop-
ular machine learning methods include neural networks and self-organizing maps
(SOMs) [75-77], kernel methods [78], especially support vector machines (SVMs)
[79,80], decision trees [81], and Bayesian methods [82], the latter being probabilis-
tic approaches for activity prediction. The computational architectures of SOMs,
decision trees, and SVMs are illustrated in Figure 15.5.
15.8.1 Self-Organizing Maps vs. Decision Trees
Neural networks and decision trees are paradigmatic examples for machine learning
approaches that have “black box” character or are interpretable in chemical terms,
respectively. This distinction is frequently considered in machine learning. Neural
networks for LBVS are derived based on learning sets of compounds and chosen
FIGURE 15.5
Representative machine learning approaches. From left to right, SOMs, deci-
sion trees, and SVMs are illustrated schematically. (
See insert for color representation of the
figure.
)
