Tailored scoring functions

The conceptually similar COMBINE approach by Wade and coworkers [134] is based on the analysis of force field energy terms per residue to describe interaction differences of a congeneric series. The predictive regression equations were reported to find application in structure-based design [135-137]. 

Another approach based on statistical analysis of residue-specific protein-ligand interaction terms is named TScore [138[. Here, individual protein-ligand interactions are evaluated for each residue within 8 A around the ligand. Force field-based descriptors are replaced by Chemscore [96] and PLP [139] terms for hydrogen bonds, lipophilicity, and steric contact. Its application to a series of factor Xa inhibitors resulted in a predictive tailored scoring function vdth 52 selected descriptors after PLS analysis q = 0.( 4, r = 0.83). This concept allows a visual interpretation of 

Other attempts are based on tailor-made empirical functions by adding descriptors and determine weights for these novel plus the known terms using appropriate statistics and a training set of several ligands. Retraining of a function using focused 

In a similar fashion, Catana and Stouten [143] developed multiple functions based on QXP protein-ligand interaction terms [112]. The concept was also applied for improving a force field-based scoring function for CDK-2 using short MD simulations based on quantum-refined force fields [144]. Here, the statistical analysis of 

The target-specific optimization method (TOP) by Seifert focuses entirely on discrimination between actives and decoys [148]. Enrichment factors were shown to be insufficient to evaluate the discrimination properties of a scoring function. 

---

Fig. 14.3 (A) Deriving a tailored scoring function using AFMoC [80], which integrates protein- and...

Matter, H., Will, D.W., Nazare, M., Schreuder, H., Laux, V., and Wehner, V. (2005) Structural requirements for factor Xa inhibition by 3-oxybenzamides with neutral PI substituents combining X-ray crystallography, 3D-QSAR and tailored scoring functions. Journal of Medicinal Chemistry, 48, 3290-3312. 

Clearly, there is still much need for improvement of scoring functions, and tailored strategies beyond the application of a single score list of a standard scoring function must be recommended for virtual screening applications. 

First we should distinguish betwe tailored and general-purpose scoring functions. When the binding affinity of a series of homologous inhibitors into a particular site is known, it is often possible to tailor a specific scoring function to fit the data. While these scoring functions are locally powerful, the methods used are typically not transferable between systems. 

---