Pharmacology profiling of chemicals and proteins

The central tenet in drug-development of one drug selectively interacts with one target is increasingly challenged by the vast amount of data released to the public domain in the past 10-15 years, documenting multiple targets for most of the FDA approved pharmaceuticals [1]. Unintended interactions between pharmaceuticals and proteins in vivo potential leads to unwanted adverse effects, toxicity and reduced half-life, but can also lead to novel therapeutic effects of already approved pharmaceuticals. Hence identification of in vivo targets is of importance in discovery, development and repurposing of pharmaceuticals, a process referred to as pharmacology profiling.
Pharmacology profiling of chemical and protein based pharmaceuticals has been proven valuable in a number studies [2], however missing values in the drug-protein interaction matrix limits the profile for novel or less studied compounds. This limitation complicates adverse effect assessment in the early drug-development phase, thus contributing to drugattrition. Prediction models offer the possibility to close these gaps and provide more complete pharmacology profiles, however improvements in performances are required for these tools to serve as an alternative to experimentally obtained measurements.
Here I present several different tools that aid pharmacology profiling of the two main classes of pharmaceuticals; chemicals (small molecules) and proteins (biopharmaceuticals).
Biopharmaceuticals have the inherent risks of eliciting an immune response due to its nonself origin, which potentially alters the pharmacology profile of the substance. The neutralization of biopharmaceuticals by antidrug antibodies (ADAs) is an important element in the immune response cascade, however studies of ADA binding site on biopharmaceuticals, referred to as B-cell epitopes, are complicated by expensive experimental procedures thus making prediction models an appealing alternative. In general, B-cell epitope prediction tools have moderate performances, which to some extent originates from an incomplete understanding of what constitute a B-cell epitope andincomplete datasets used for model building and benchmarking. In the first paper included in this thesis we analyze the B-cell epitopes obtained from co-chrystalized antibody-antigen complexes from the PDB database. We were able to describe the epitope area as a flat oblong area residing on the protein surface consisting of a hydrophobic core surrounded by hydrophilic/charged residues.
This finding prompted us to update the B-cell epitope prediction method DiscoTope [3] by introducing a novel scoring function for describing the spatial neighborhood surrounding each residue as described in paper II of this thesis. Using the developed method we assessed the impact on performance using a more realistic benchmark definition compared to privies studies, by including multiple epitopes for each antigen and the biological unit used for raising the antibody response, when available. On average, the Area Under the roc-Curve(AUC) performance was improved from 0.791 to 0.824 for the 13 proteins were additional information could be obtained, thereby indicating that the performance of B-cell epitope prediction tools in general are under-estimated.
Novel techniques such as Next Generation Sequencing (NGS) and peptide microarray facilitate novel strategies for experimental identification of B-cell epitopes. In chapter 4, a novel method for epitope identification is described, combining NGS with phage-display.Epitopes in peanut allergen ara h1 were successfully detected using sera from peanut allergic patients and confirmed using peptide micro-array technology, demonstrating the applicability of both methods.
Adverse effect of small molecule based pharmaceutical is rarely mediated through an immune response but is predominantly the consequence of interactions with unintended proteins in vivo. To assists researchers in determine the binding profile of chemicals, thus their pharmacology profile, a database of chemical-protein interactions were developed and presented in chapter 5. The database integrates chemical-protein interaction information from 10 different databases as well as disease, functional and pathway mapping of proteins, SNP data through the Ensembl database and prediction tools for filling out gaps in the chemical-protein interaction matrix. Graphical representation of the pharmacology space is accomplished by the use of zoomable heatmaps, which enable traversing from an overview of the entire space to specific pharmacology profiles of a single chemical by zooming on specific areas of the heatmap. The compiled dataset together with the implemented visualization and prediction tools, facilitate pharmacology profiling of chemicals in all development stages, hence potentially enable identification of adverse effects in early drugdevelopmentor identification of novel treatment paradigms for approved pharmaceuticals.
Finally, the visualization of the pharmacology space is addressed by developing a 2 dimensional zoomable heatmap inspired by country and city maps. Chemicals sharing similar scaffold or features are placed together on the map, thus enable a more detailed visualization of the pharmacology landscape surrounding one or more chemicals of interest. The tool, presented in chapter 6, enables researchers to couple scaffold and feature hopping with bioactivity data for the use in drug-discovery and development, thus avoiding unwanted adverse effects.
In summary, here I present several different tools that can assists researchers in determine essential properties in the pharmacology profile of both protein and small molecule pharmaceuticals and potentially detect adverse effects in drug-development.