Structure-based identification of small molecule binding sites using a free energy model

We separately have shown that the maximal druglike affinity of a given binding site on a protein can be calculated on the basis of the binding-site structure alone by using a desolvation-based free energy model along with the notion that druglike ligands fall into certain physiochemical property ranges. Here, we present an approach where we reformulate the calculated druggability affinity as an additive free energy to facilitate the searching of whole protein surfaces for druglike binding sites. The highest-scoring patches in many cases represent known ligand-binding sites for druggable targets, but not for difficult targets. This approach differs from other approaches in that it does not simply identify pockets with the greatest volume but instead identifies pockets that are likely to be amenable to druglike small-molecule binding. Combining the method with a functional residue prediction method called SCA (statistical coupling analysis) results in the prediction of potentially druggable allosteric binding sites on p38alpha kinase.     

Drug-like density: a method of quantifying the "bindability" of a protein target based on a very large set of pockets and drug-like ligands from the Protein Data Bank

One approach to estimating the "chemical tractability" of a candidate protein target where we know the atomic resolution structure is to examine the physical properties of potential binding sites. A number of other workers have addressed this issue. We characterize ~290,000 "pockets" from ~42,000 protein crystal structures in terms of a three parameter "pocket space": volume, buriedness, and hydrophobicity. A metric DLID (drug-like density) measures how likely a pocket is to bind a drug-like molecule. This is calculated from the count of other pockets in its local neighborhood in pocket space that contain drug-like cocrystallized ligands and the count of total pockets in the neighborhood. Surprisingly, despite being defined locally, a global trend in DLID can be predicted by a simple linear regression on log(volume), buriedness, and hydrophobicity. Two levels of simplification are necessary to relate the DLID of individual pockets to "targets": taking the best DLID per Protein Data Bank (PDB) entry (because any given crystal structure can have many pockets), and taking the median DLID over all PDB entries for the same target (because different crystal structures of the same protein can vary because of artifacts and real conformational changes). We can show that median DLIDs for targets that are detectably homologous in sequence are reasonably similar and that median DLIDs correlate with the "druggability" estimate of Cheng et al. (Nature Biotechnology 2007, 25, 71-75).     

DrugPred: a structure-based approach to predict protein druggability developed using an extensive nonredundant data set

Judging if a protein is able to bind orally available molecules with high affinity, i.e. if a protein is druggable, is an important step in target assessment. In order to derive a structure-based method to predict protein druggability, a comprehensive, nonredundant data set containing crystal structures of 71 druggable and 44 less druggable proteins was compiled by literature search and data mining. This data set was subsequently used to train a structure-based druggability predictor (DrugPred) using partial least-squares projection to latent structures discriminant analysis (PLS-DA). DrugPred performed well in discriminating druggable from less druggable binding sites for both internal and external predictions. The method is robust against conformational changes in the binding site and outperforms previously published methods. The superior performance of DrugPred is likely due to the size and composition of the training set which, in contrast to most previously developed methods, only contains cavities that have evolved to bind a natural ligand.     

支持向量机与KStar模型预测细胞色素P450酶催化的氧脱烃反应

分别以支持向量机(SVM)和KStar方法为基础, 构建了代谢产物的分子形状判别和代谢反应位点判别的嵌套预测模型. 分子形状判别模型是以272个分子为研究对象, 计算了包括分子拓扑、二维自相关、几何结构等在内的1280个分子描述符, 考查了支持向量机、决策树、贝叶斯网络、k最近邻这四种机器学习方法建立分类预测模型的准确性. 结果表明, 支持向量机优于其他方法, 此模型可用于预测分子能否被细胞色素P450酶催化发生氧脱烃反应. 代谢反应位点判别模型以538个氧脱烃反应代谢位点为研究对象, 计算了表征原子能量、价态、电荷等26个量子化学特征, 比较了决策树、贝叶斯网络、KStar、人工神经网络建模的准确率. 结果显示, KStar模型的准确率、敏感性、专一性均在90%以上, 对分子形状判别模型筛选出的分子, 此模型能较好地判断出哪个C―O键发生断裂. 本文以15个代谢反应明确的中药分子为验证集, 验证模型准确性, 研究结果表明基于SVM和KStar的嵌套预测模型具有一定的准确性, 有助于开展中药分子氧脱烃代谢产物的预测研究.     

Druggability Assessment of Allosteric Proteins by Dynamics Simulations in the Presence of Probe Molecules

Druggability assessment of a target protein has emerged in recent years as an important concept in hit-to-lead optimization. A reliable and physically relevant measure of druggability would allow informed decisions on the risk of investing in a particular target. Here, we define "druggability" as a quantitative estimate of binding sites and affinities for a potential drug acting on a specific protein target. In the present study, we describe a new methodology that successfully predicts the druggability and maximal binding affinity for a series of challenging targets, including those that function through allosteric mechanisms. Two distinguishing features of the methodology are (i) simulation of the binding dynamics of a diversity of probe molecules selected on the basis of an analysis of approved drugs and (ii) identification of druggable sites and estimation of corresponding binding affinities on the basis of an evaluation of the geometry and energetics of bound probe clusters. The use of the methodology for a variety of targets such as murine double mutant-2, protein tyrosine phosphatase 1B (PTP1B), lymphocyte function-associated antigen 1, vertebrate kinesin-5 (Eg5), and p38 mitogen-activated protein kinase provides examples for which the method correctly captures the location and binding affinities of known drugs. It also provides insights into novel druggable sites and the target's structural changes that would accommodate, if not promote and stabilize, drug binding. Notably, the ability to identify high affinity spots even in challenging cases such as PTP1B or Eg5 shows promise as a rational tool for assessing the druggability of protein targets and identifying allosteric or novel sites for drug binding.     

Druggability of methyl-lysine binding sites

Structural modules that specifically recognize—or read—methylated or acetylated lysine residues on histone peptides are important components of chromatin-mediated signaling and epigenetic regulation of gene expression. Deregulation of epigenetic mechanisms is associated with disease conditions, and antagonists of acetyl-lysine binding bromodomains are efficacious in animal models of cancer and inflammation, but little is known regarding the druggability of methyl-lysine binding modules. We conducted a systematic structural analysis of readers of methyl marks and derived a predictive druggability landscape of methyl-lysine binding modules. We show that these target classes are generally less druggable than bromodomains, but that some proteins stand as notable exceptions.     

Inhibition of protein interactions: co-crystalized protein–protein interfaces are nearly as good as holo proteins in rigid-body ligand docking

Modulating protein interaction pathways may lead to the cure of many diseases. Known protein–protein inhibitors bind to large pockets on the protein–protein interface. Such large pockets are detected also in the protein–protein complexes without known inhibitors, making such complexes potentially druggable. The inhibitor-binding site is primary defined by the side chains that form the largest pocket in the protein-bound conformation. Low-resolution ligand docking shows that the success rate for the protein-bound conformation is close to the one for the ligand-bound conformation, and significantly higher than for the apo conformation. The conformational change on the protein interface upon binding to the other protein results in a pocket employed by the ligand when it binds to that interface. This proof-of-concept study suggests that rather than using computational pocket-opening procedures, one can opt for an experimentally determined structure of the target co-crystallized protein–protein complex as a starting point for drug design.     

基于三维分子描述符的埃坡霉素类衍生物抗癌活性预测

为预测埃坡霉素类衍生物的抗癌活性, 定义了一套表征分子形状的描述符, 即K阶形状参数, 并计算了67个表征分子的电子、拓扑和几何结构的分子描述符. 描述符经遗传算法筛选, 用于建立基于支持向量学习机(SVM)的抗癌活性分类模型; 用留一法和5重交叉验证法对SVM模型参数进行了优化. 结果表明模型具有较高的预测性且两种方法得到相近结果, 交叉验证的预测正确率达80.6%; 经筛选后的描述符有30个, 其中含有5个K阶形状参数, 这些描述符对埃坡霉素类衍生物的抗癌活性的模型建立具有比较重要的作用.     

Druggable Allosteric Sites in β-Propeller Lectins

Carbohydrate-binding proteins (lectins) are auspicious targets in drug discovery to combat antimicrobial resistance; however, their non-carbohydrate drug-like inhibitors are still unavailable. Here, we present a druggable pocket in a β-propeller lectin BambL from Burkholderia ambifaria as a potential target for allosteric inhibitors. This site was identified employing ¹⁹F NMR fragment screening and a computational pocket prediction algorithm SiteMap. The structure–activity relationship study revealed the most promising fragment with a dissociation constant of 0.3±0.1 mM and a ligand efficiency of 0.3 kcal mol^?1 HA^?1 that affected the orthosteric site. This effect was substantiated by site-directed mutagenesis in the orthosteric and secondary pockets. Future drug-discovery campaigns that aim to develop small molecule inhibitors can benefit from allosteric sites in lectins as a new therapeutic approach against antibiotic-resistant pathogens.