HERG computational model

Numerous structural classes of potassium ion channels have been discovered. The six transmembrane domain proteins include voltage-gated channels (Kv), and Ca2+ activated K+ channels (Kca). However, there are also proteins with 2 transmembrane domains and 7 transmembrane domains [3]. The most widely studied potassium channel of relevance to toxicity and therefore the interests of the pharmaceutical industry to date appears to be Kv 11.1, the human ether-a-go-go-related gene (hERG), as is described in more detail below. Some computational modeling of other potassium channels is also addressed. 

Indeed, considering the latter 3D QSAR model, the features that make a molecule suitable to bind to the hERG channel start delineating in a chemically interpretable manner, but, it is rather dear how these kinds of models emphasize mostly the 3D steric aspects of molecules, depending mainly on factors such as the conformation (or the conformational analysis protocol) or the alignment of the molecules. To obtain a description of the characteristics of hERG-blocking molecules in terms of measurable (computable) properties in a way that the physicochemical determinants of the activity can be identified, the classical 2D QSAR approach is well suited. 

Multivariate models using neural networks, support vector machines and least median squares regression have been used to predict hERG activity [96-98]. These types of models function more as computational black box assays. 

Recently predictive in silico modeling for hERG channel blockers has been described." Different approaches have aimed primarily at fdtering out potential hERG channel blockers in the context of combinatorial and virtual libraries and to elucidate structure-activity relationships. These new computational methods may predict trends, but are not as yet sufficiently precise to make valid predictions. 

Masetti M, Cavalli A, Recanatini M (2008) Modeling the hERG potassium channel in a phospholipid bilayer molecular dynamics and drug docking studies. J Comput Chem 29(5) 795-808... 

Fig. 6. Chemical feature similarities between whole activity classes for the targets used in predinical safety profiling. Black colors indicate high class similarities, whereas grey colors show low similarities. We compute similarity by using the Pearson correlation of normalized feature probabilities in each Bayesian model. Notably, receptor families share ligand similarity, as would be expected however, non-family pairing also share features, such as antihistamines (H1 receptor) and hERG blockers (in agreement with the arrhythmia that can be caused by this class of compounds).

Diller, D.J. (2009) In silica hERG modeling challenges and progress. Current Computer-Aided Drug Design, 5, 106-121. 

Due to the aforementioned discrepancy in data availability (especially relevant to translation of toxic effect) and the fact that many clinical endpoints are multi-mechanistic, it is important to stress that each computational step should be well defined and model small steps, for example, a traditional quantitative structure-activity relationship (QSAR) approach based on chemical structure is probably relevant to distinguish hERG binders from nonbinders, but not relevant to model a small set of diverse compounds associated with a complex endpoint such as drug induced liver injury (DILI). A second important factor to consider when construchng in silico safety models is the intended use of the model, and the potential cost associated with false positives versus false negatives from the model. For instance, there is zero... 

---