Automated design of ligands to polypharmacological profiles.

The clinical efficacy and safety of a drug is determined by its activity profile across many proteins in the proteome. However, designing drugs with a specific multi-target profile is both complex and difficult. Therefore methods to design drugs rationally a priori against profiles of several proteins would have immense value in drug discovery. Here we describe a new approach for the automated design of ligands against profiles of multiple drug targets. The method is demonstrated by the evolution of an approved acetylcholinesterase inhibitor drug into brain-penetrable ligands with either specific polypharmacology or exquisite selectivity profiles for G-protein-coupled receptors. Overall, 800 ligand-target predictions of prospectively designed ligands were tested experimentally, of which 75% were confirmed to be correct. We also demonstrate target engagement in vivo. The approach can be a useful source of drug leads when multi-target profiles are required to achieve either selectivity over other drug targets or a desired polypharmacology.

Structural bioinformatics mutation analysis reveals genotype-phenotype correlations in von Hippel-Lindau disease and suggests molecular mechanisms of tumorigenesis.

Mutations in the VHL gene lead to von Hippel-Lindau (VHL) disease, a clinically heterogeneous cancer syndrome. Here, we use software and database tools to understand and predict the phenotypes associated with missense mutations in the VHL gene product, pVHL. The protein product pVHL is known to interact with elongin B, elongin C, and the HIF substrate. By analyzing known and predicted interaction sites and predictions of thermodynamic stability change upon mutation, we generate new hypotheses regarding the molecular etiology of renal cell carcinoma (RCC) and pheochromocytoma (PCC) in VHL disease. We find that the molecular causes of RCC and PCC appear to be decoupled. RCC may arise through two distinct mechanisms: disruption of HIF interactions or binding at the elongin B interface. PCC is triggered by mutations which disrupt interactions at the elongin C binding site. These findings have important implications for VHL disease and for nonfamilial RCC, because most cases of clear cell RCC are linked with VHL inactivation. Additionally, predicting effects of genetic variation will be critical as genetic sequencing accelerates; the analytical strategy presented here may elucidate other systems as further data on genetic variation become available.

Structural and functional restraints in the evolution of protein families and superfamilies.

Divergent evolution of proteins reflects both selectively advantageous and neutral amino acid substitutions. In the present article, we examine restraints on sequence, which arise from selectively advantageous roles for structure and function and which lead to the conservation of local sequences and structures in families and superfamilies. We analyse structurally aligned members of protein families and superfamilies in order to investigate the importance of the local structural environment of amino acid residues in the acceptance of amino acid substitutions during protein evolution. We show that solvent accessibility is the most important determinant, followed by the existence of hydrogen bonds from the side-chain to main-chain functions and the nature of the element of secondary structure to which the amino acid contributes. Polar side chains whose hydrogen-bonding potential is satisfied tend to be more conserved than their unsatisfied or non-hydrogen-bonded counterparts, and buried and satisfied polar residues tend to be significantly more conserved than buried hydrophobic residues. Finally, we discuss the importance of functional restraints in the form of interactions of proteins with other macromolecules in assemblies or with substrates, ligands or allosteric regulators. We show that residues involved in such functional interactions are significantly more conserved and have differing amino acid substitution patterns.

Meta-QSAR: a large-scale application of meta-learning to drug design and discovery.

We investigate the learning of quantitative structure activity relationships (QSARs) as a case-study of meta-learning. This application area is of the highest societal importance, as it is a key step in the development of new medicines. The standard QSAR learning problem is: given a target (usually a protein) and a set of chemical compounds (small molecules) with associated bioactivities (e.g. inhibition of the target), learn a predictive mapping from molecular representation to activity. Although almost every type of machine learning method has been applied to QSAR learning there is no agreed single best way of learning QSARs, and therefore the problem area is well-suited to meta-learning. We first carried out the most comprehensive ever comparison of machine learning methods for QSAR learning: 18 regression methods, 3 molecular representations, applied to more than 2700 QSAR problems. (These results have been made publicly available on OpenML and represent a valuable resource for testing novel meta-learning methods.) We then investigated the utility of algorithm selection for QSAR problems. We found that this meta-learning approach outperformed the best individual QSAR learning method (random forests using a molecular fingerprint representation) by up to 13%, on average. We conclude that meta-learning outperforms base-learning methods for QSAR learning, and as this investigation is one of the most extensive ever comparisons of base and meta-learning methods ever made, it provides evidence for the general effectiveness of meta-learning over base-learning.

A structural bioinformatics approach to the analysis of nonsynonymous single nucleotide polymorphisms (nsSNPs) and their relation to disease.

The prediction of the effects of nonsynonymous single nucleotide polymorphisms (nsSNPs) on function depends critically on exploiting all information available on the three-dimensional structures of proteins. We describe software and databases for the analysis of nsSNPs that allow a user to move from SNP to sequence to structure to function. In both structure prediction and the analysis of the effects of nsSNPs, we exploit information about protein evolution, in particular, that derived from investigations on the relation of sequence to structure gained from the study of amino acid substitutions in divergent evolution. The techniques developed in our laboratory have allowed fast and automated sequence-structure homology recognition to identify templates and to perform comparative modeling; as well as simple, robust, and generally applicable algorithms to assess the likely impact of amino acid substitutions on structure and interactions. We describe our strategy for approaching the relationship between SNPs and disease, and the results of benchmarking our approach -- human proteins of known structure and recognized mutation.

Quantifying the chemical beauty of drugs.

Drug-likeness is a key consideration when selecting compounds during the early stages of drug discovery. However, evaluation of drug-likeness in absolute terms does not reflect adequately the whole spectrum of compound quality. More worryingly, widely used rules may inadvertently foster undesirable molecular property inflation as they permit the encroachment of rule-compliant compounds towards their boundaries. We propose a measure of drug-likeness based on the concept of desirability called the quantitative estimate of drug-likeness (QED). The empirical rationale of QED reflects the underlying distribution of molecular properties. QED is intuitive, transparent, straightforward to implement in many practical settings and allows compounds to be ranked by their relative merit. We extended the utility of QED by applying it to the problem of molecular target druggability assessment by prioritizing a large set of published bioactive compounds. The measure may also capture the abstract notion of aesthetics in medicinal chemistry.

What can we learn from the evolution of protein-ligand interactions to aid the design of new therapeutics?

Efforts to increase affinity in the design of new therapeutic molecules have tended to lead to greater lipophilicity, a factor that is generally agreed to be contributing to the low success rate of new drug candidates. Our aim is to provide a structural perspective to the study of lipophilic efficiency and to compare molecular interactions created over evolutionary time with those designed by humans. We show that natural complexes typically engage in more polar contacts than synthetic molecules bound to proteins. The synthetic molecules also have a higher proportion of unmatched heteroatoms at the interface than the natural sets. These observations suggest that there are lessons to be learnt from Nature, which could help us to improve the characteristics of man-made molecules. In particular, it is possible to increase the density of polar contacts without increasing lipophilicity and this is best achieved early in discovery while molecules remain relatively small.

Rapid analysis of pharmacology for infectious diseases.

Pandemic, epidemic and endemic infectious diseases are united by a common problem: how do we rapidly and cost-effectively identify potential pharmacological interventions to treat infections? Given the large number of emerging and neglected infectious diseases and the fact that they disproportionately afflict the poorest members of the global society, new ways of thinking are required to developed high productivity discovery systems that can be applied to a larger number of pathogens. The growing availability of parasite genome data provides the basis for developing methods to prioritize, a priori, the potential drug target and pharmacological landscape of an infectious disease. Thus the overall objective of infectious disease informatics is to enable the rapid generation of plausible, novel medical hypotheses of testable pharmacological experiments, by uncovering undiscovered relationships in the wealth of biomedical literature and databases that were collected for other purposes. In particular our goal is to identify potential drug targets present in a pathogen genome and prioritize which pharmacological experiments are most likely to discover drug-like lead compounds rapidly against a pathogen (i.e. which specific compounds and drug targets should be screened, in which assays and where they can be sourced). An integral part of the challenge is the development and integration of methods to predict druggability, essentiality, synthetic lethality and polypharmacology in pathogen genomes, while simultaneously integrating the inevitable issues of chemical tractability and the potential for acquired drug resistance from the start.

An ontology for description of drug discovery investigations.

The paper presents an ontology for the description of Drug Discovery Investigation (DDI).This has been developed through the use of a Robot Scientist "Eve", and in consultation with industry. DDI aims to define the principle entities and the relations in the research and development phase of the drug discovery pipeline. DDI is highly transferable and extendable due to its adherence to accepted standards, and compliance with existing ontology resources. This enables DDI to be integrated with such related ontologies as the Vaccine Ontology, the Advancing Clinico-Genomic Trials on Cancer Master Ontology, etc. DDI is available at http://purl.org/ddi/wikipedia or http://purl.org/ddi/home.

Atomic interactions and profile of small molecules disrupting protein-protein interfaces: the TIMBAL database.

Growing evidence of the possibility of modulating protein-protein interactions with small molecules is opening the door to new approaches and concepts in drug discovery. In this paper, we describe the creation of TIMBAL, a hand-curated database holding an up to date collection of small molecules inhibiting multi-protein complexes. This database has been analysed and profiled in terms of molecular properties. Protein-protein modulators tend to be large lipophilic molecules with few hydrogen bond features. An analysis of TIMBAL's intersection with other structural databases, including CREDO (protein-small molecule from the PDB) and PICCOLO (protein-protein from the PDB) reveals that TIMBAL molecules tend to form mainly hydrophobic interactions with only a few hydrogen bonding contacts. With respect to potency, TIMBAL molecules are slightly less efficient than an average medicinal chemistry hit or lead. The database provides a resource that will allow further insights into the types of molecules favoured by protein interfaces and provide a background to continuing work in this area. Access at http://www-cryst.bioc.cam.ac.uk/timbal.