Datasets, for virtual screening

Bigure 16.5 summarizes our approach to using validated QSAR models for virtual screening as applied to the anticonvulsant dataset. It presents a practical example of the drug discovery workflow that can be generalized for any dataset in which sufficient data to develop reliable QSAR models is available. 

Figure 5.8 Distribution of the predicted values of selectivity and activity to the A3 adenosine receptor for the filtered virtual screening dataset.

Figure 5.10 Distribution of the predicted values of affinity and intrinsic activity to the MTi and MT2 melatonin receptors for the filtered virtual screening dataset of indole derivatives, (a) MTi affinity MTi intrinsic activity, b) MT2 affinity vi. MT2 intrinsic activity, (c) MTi MT2 affinity, d) MTi Vi. MT2 intrinsic activity.

Many different descriptors have been developed in an attempt to define chemistry spaces that are relevant for bioactivity. Aside from being relevant to biological activity, to be useful for combinatorial chemistry, descriptors should also be relatively easy to calculate in order that the methods can be applied to large datasets. Some descriptors commonly used in virtual screening are described below. For more detail on descriptors, see the chapter by Downs and Barnard and the earlier review by Brown [9]. The most commonly used similarity coefficients are the Tanimoto coefficient and Euclidean distance [10]. A weighting scheme may also be applied to assign relative weights to different descriptors. 

---