Properties of protein drug target classes.

Accurate identification of drug targets is a crucial part of any drug development program. We mined the human proteome to discover properties of proteins that may be important in determining their suitability for pharmaceutical modulation. Data was gathered concerning each protein's sequence, post-translational modifications, secondary structure, germline variants, expression profile and drug target status. The data was then analysed to determine features for which the target and non-target proteins had significantly different values. This analysis was repeated for subsets of the proteome consisting of all G-protein coupled receptors, ion channels, kinases and proteases, as well as proteins that are implicated in cancer. Machine learning was used to quantify the proteins in each dataset in terms of their potential to serve as a drug target. This was accomplished by first inducing a random forest that could distinguish between its targets and non-targets, and then using the random forest to quantify the drug target likeness of the non-targets. The properties that can best differentiate targets from non-targets were primarily those that are directly related to a protein's sequence (e.g. secondary structure). Germline variants, expression levels and interactions between proteins had minimal discriminative power. Overall, the best indicators of drug target likeness were found to be the proteins' hydrophobicities, in vivo half-lives, propensity for being membrane bound and the fraction of non-polar amino acids in their sequences. In terms of predicting potential targets, datasets of proteases, ion channels and cancer proteins were able to induce random forests that were highly capable of distinguishing between targets and non-targets. The non-target proteins predicted to be targets by these random forests comprise the set of the most suitable potential future drug targets, and should therefore be prioritised when building a drug development programme.

Maximising the size of non-redundant protein datasets using graph theory.

Analysis of protein data sets often requires prior removal of redundancy, so that data is not biased by containing similar proteins. This is usually achieved by pairwise comparison of sequences, followed by purging so that no two pairs have similarities above a chosen threshold. From a starting set, such as the PDB or a genome, one should remove as few sequences as possible, to give the largest possible non-redundant set for subsequent analysis. Protein redundancy can be represented as a graph, with proteins as nodes connected by undirected edges, if they have a pairwise similarity above the chosen threshold. The problem is then equivalent to finding the maximum independent set (MIS), where as few nodes are removed as possible to remove all edges. We tested seven MIS algorithms, three of which are new. We applied the methods to the PDB, subsets of the PDB, various genomes and the BHOLSIB benchmark datasets. For PDB subsets of up to 1000 proteins, we could compare to the exact MIS, found by the Cliquer algorithm. The best algorithm was the new method, Leaf. This works by adding clique members that have no edges to nodes outside the clique to the MIS, starting with the smallest cliques. For PDB subsets of up to 1000 members, it usually finds the MIS and is fast enough to apply to data sets of tens of thousands of proteins. Leaf gives sets that are around 10% larger than the commonly used PISCES algorithm, that are of identical quality. We therefore suggest that Leaf should be the method of choice for generating non-redundant protein data sets, though it is ineffective on dense graphs, such as the BHOLSIB benchmarks. The Leaf algorithm is available at: https://github.com/SimonCB765/Leaf, and sets from genomes and the PDB are available at: http://www.bioinf.manchester.ac.uk/leaf/.