James L. Melville

Machine Learning in Virtual Screening

In this review, we highlight recent applications of machine learning to virtual screening, focusing on the use of supervised techniques to train statistical learning algorithms to prioritize databases of molecules as active against a particular protein target. Both ligand-based similarity searching and structure-based docking have benefited from machine learning algorithms, including naïve Bayesian classifiers, support vector machines, neural networks, and decision trees, as well as more traditional regression techniques. Effective application of these methodologies requires an appreciation of data preparation, validation, optimization, and search methodologies, and we also survey developments in these areas.

Contemporary QSAR Classifiers Compared

We present a comparative assessment of several state-of-the-art machine learning tools for mining drug data, including support vector machines (SVMs) and the ensemble decision tree methods boosting, bagging, and random forest, using eight data sets and two sets of descriptors. We demonstrate, by rigorous multiple comparison statistical tests, that these techniques can provide consistent improvements in predictive performance over single decision trees. However, within these methods, there is no clearly best-performing algorithm. This motivates a more in-depth investigation into the properties of random forests. We identify a set of parameters for the random forest that provide optimal performance across all the studied data sets. Additionally, the tree ensemble structure of the forest may provide an interpretable model, a considerable advantage over SVMs. We test this possibility and compare it with standard decision tree models.

FieldScreen: virtual screening using molecular fields. Application to the DUD data set.

FieldScreen, a ligand-based Virtual Screening (VS) method, is described. Its use of 3D molecular fields makes it particularly suitable for scaffold hopping, and we have rigorously validated it for this purpose using a clustered version of the Directory of Useful Decoys (DUD). Using thirteen pharmaceutically relevant targets, we demonstrate that FieldScreen produces superior early chemotype enrichments, compared to DOCK. Additionally, hits retrieved by FieldScreen are consistently lower in molecular weight than those retrieved by docking. Where no X-ray protein structures are available, FieldScreen searches are more robust than docking into homology models or apo structures.

Novel lead structures for p38 MAP kinase via FieldScreen virtual screening.

p38 MAP kinase has received considerable interest in the pharmaceutical industry and remains a valid and interesting target for the treatment of inflammation. To discover novel p38 inhibitors, we applied the ligand-based virtual screening technique, FieldScreen, to 1.2 million commercially available compounds. Fifty-eight diverse compounds were selected for biological analysis, using molecular field similarity to known inhibitors, while explicitly removing any structure that shared a scaffold with previously reported p38 inhibitors. Of these, 11 (19%) showed >or=20% inhibition of p38 at 10 microM. We chose to prepare analogues of two distinct chemical series resulting in a potential lead compound with pIC(50) of 6.4. Modeling of SAR using FieldAlign, a ligand alignment protocol, was used to rationalize the SAR of the series of thiadiazole based inhibitors.

TMACC: interpretable correlation descriptors for quantitative structure-activity relationships.

Highly predictive topological maximum cross correlation (TMACC) descriptors for the derivation of quantitative structure-activity relationships (QSARs) are presented, based on the widely used autocorrelation method. They require neither the calculation of three-dimensional conformations nor an alignment of structures. We have validated the TMACC descriptors across eight literature data sets, ranging in size from 66 to 361 molecules. In combination with partial least-squares regression, they perform competitively with a current state-of-the-art 2D QSAR methodology, hologram QSAR (HQSAR), yielding larger leave-one-out cross-validated coefficient of determination values (LOO q2) for five data sets. Like HQSAR, these descriptors are also interpretable but do not require hashing. The interpretation both enables the automated extraction of SARs and can give a description in qualitative agreement with more time-consuming 3D and 4D QSAR methods. Open source software for generating the TMACC descriptors is freely available from our Web site.

Molecular docking and QSAR of aplyronine A and analogues: potent inhibitors of actin

Actin-binding natural products have been identified as a potential basis for the design of cancer therapeutic agents. We report flexible docking and QSAR studies on aplyronine A analogues. Our findings show the macrolide ‘tail’ to be fundamental for the depolymerisation effect of actin-binding macrolides and that it is the tail which forms the initial interaction with the actin rather than the macrocycle, as previously believed. Docking energy scores for the compounds were highly correlated with actin depolymerisation activity. The 3D-QSAR models were predictive, with the best model giving a q2 value of 0.85 and a r2 of 0.94. Results from the docking simulations and the interpretation from QSAR “coeff*stdev” contour maps provide insight into the binding mechanism of each analogue and highlight key features that influence depolymerisation activity. The results herein may aid the design of a putative set of analogues that can help produce efficacious and tolerable anti-tumour agents. Finally, using the best QSAR model, we have also made genuine predictions for an independent set of recently reported aplyronine analogues.

Similarity by Compression

We present a simple and effective method for similarity searching in virtual high-throughput screening, requiring only a string-based representation of the molecules (e.g., SMILES) and standard compression software, available on all modern desktop computers. This method utilizes the normalized compression distance, an approximation of the normalized information distance, based on the concept of Kolmogorov complexity. On representative data sets, we demonstrate that compression-based similarity searching can outperform standard similarity searching protocols, exemplified by the Tanimoto coefficient combined with a binary fingerprint representation and data fusion. Software to carry out compression-based similarity is available from our Web site at http://comp.chem.nottingham.ac.uk/download/zippity.

Computational screening of combinatorial catalyst libraries.

A catalyst design methodology, utilizing combinatorial synthesis in parallel with chemometric analysis, is presented, which considers the 3D steric and electrostatic properties of substituents about a constant core structure.