Coverage Score: A Model Agnostic Method to Efficiently Explore Chemical Space.

Selecting the most appropriate compounds to synthesize and test is a vital aspect of drug discovery. Methods like clustering and diversity present weaknesses in selecting the optimal sets for information gain. Active learning techniques often rely on an initial model and computationally expensive semi-supervised batch selection. Herein, we describe a new subset-based selection method, Coverage Score, that combines Bayesian statistics and information entropy to balance representation and diversity to select a maximally informative subset. Coverage Score can be influenced by prior selections and desirable properties. In this paper, subsets selected through Coverage Score are compared against subsets selected through model-independent and model-dependent techniques for several datasets. In drug-like chemical space, Coverage Score consistently selects subsets that lead to more accurate predictions compared to other selection methods. Subsets selected through Coverage Score produced Random Forest models that have a root-mean-square-error up to 12.8% lower than subsets selected at random and can retain up to 99% of the structural dissimilarity of a diversity selection.

Design of a multi-purpose fragment screening library using molecular complexity and orthogonal diversity metrics.

Fragment Based Drug Discovery (FBDD) continues to advance as an efficient and alternative screening paradigm for the identification and optimization of novel chemical matter. To enable FBDD across a wide range of pharmaceutical targets, a fragment screening library is required to be chemically diverse and synthetically expandable to enable critical decision making for chemical follow-up and assessing new target druggability. In this manuscript, the Pfizer fragment library design strategy which utilized multiple and orthogonal metrics to incorporate structure, pharmacophore and pharmacological space diversity is described. Appropriate measures of molecular complexity were also employed to maximize the probability of detection of fragment hits using a variety of biophysical and biochemical screening methods. In addition, structural integrity, purity, solubility, fragment and analog availability as well as cost were important considerations in the selection process. Preliminary analysis of primary screening results for 13 targets using NMR Saturation Transfer Difference (STD) indicates the identification of uM-mM hits and the uniqueness of hits at weak binding affinities for these targets.

An Open Drug Discovery Competition: Experimental Validation of Predictive Models in a Series of Novel Antimalarials.

The Open Source Malaria (OSM) consortium is developing compounds that kill the human malaria parasite, Plasmodium falciparum, by targeting PfATP4, an essential ion pump on the parasite surface. The structure of PfATP4 has not been determined. Here, we describe a public competition created to develop a predictive model for the identification of PfATP4 inhibitors, thereby reducing project costs associated with the synthesis of inactive compounds. Competition participants could see all entries as they were submitted. In the final round, featuring private sector entrants specializing in machine learning methods, the best-performing models were used to predict novel inhibitors, of which several were synthesized and evaluated against the parasite. Half possessed biological activity, with one featuring a motif that the human chemists familiar with this series would have dismissed as "ill-advised". Since all data and participant interactions remain in the public domain, this research project "lives" and may be improved by others.

Searching chemical space with the Bayesian Idea Generator.

The Pfizer Global Virtual Library (PGVL) is defined as a set compounds that could be synthesized using validated protocols and monomers. However, it is too large (10(12) compounds) to search by brute-force methods for close analogues of a given input structure. In this paper the Bayesian Idea Generator is described which is based on a novel application of Bayesian statistics to narrow down the search space to a prioritized set of existing library arrays (the default is 16). For each of these libraries the 6 closest neighbors are retrieved from the existing compound file, resulting in a screenable hypothesis of 96 compounds. Using the Bayesian models for library space, the Pfizer file of singleton compounds has been mapped to library space and is optionally searched as well. The method is >99% accurate in retrieving known library provenance from an independent test set. The compounds retrieved strike a balance between similarity and diversity resulting in frequent scaffold hops. Four examples of how the Bayesian Idea Generator has been successfully used in drug discovery are provided. The methodology of the Bayesian Idea Generator can be used for any collection of compounds containing distinct clusters, and an example using compound vendor catalogues has been included.

Global mapping of pharmacological space.

We present the global mapping of pharmacological space by the integration of several vast sources of medicinal chemistry structure-activity relationships (SAR) data. Our comprehensive mapping of pharmacological space enables us to identify confidently the human targets for which chemical tools and drugs have been discovered to date. The integration of SAR data from diverse sources by unique canonical chemical structure, protein sequence and disease indication enables the construction of a ligand-target matrix to explore the global relationships between chemical structure and biological targets. Using the data matrix, we are able to catalog the links between proteins in chemical space as a polypharmacology interaction network. We demonstrate that probabilistic models can be used to predict pharmacology from a large knowledge base. The relationships between proteins, chemical structures and drug-like properties provide a framework for developing a probabilistic approach to drug discovery that can be exploited to increase research productivity.

The ELF Honest Data Broker: informatics enabling public-private collaboration in a precompetitive arena.

New precompetitive ways of working in the pharmaceutical industry are driving the development of new informatics systems to enable their execution and management. The European Lead Factory (ELF) is a precompetitive, 30-partner collaboration between academic groups, small-medium enterprises and pharmaceutical companies created to discover small molecule hits against novel biological targets. A unique HTS screening and triage workflow has been developed to balance the intellectual property and scientific requirements of all the partners. Here, we describe the ELF Honest Data Broker, a cloud-based informatics system providing the scientific triage tools, fine-grained permissions and management tools required to implement the workflow.

Identification of a second binding site in the estrogen receptor.

Fluorescence spectrometry data by Tyulmenkov and Klinge (Arch. Biochem. Biophys. 2000, 381, 135-142) suggest the presence of a second binding site in both subtypes ER alpha and ER beta of the estrogen receptor (ER). A cavity previously described as a solvent channel was located in close proximity to the steroid binding site of both ER subtypes. Derivatives of a tetrahydrochrysene (THC) compound, speculated in the literature to bind to a second binding site, were docked successfully in the second sites identified. However, computation of accurate interaction scores indicates preferred binding to the steroid binding site over the second binding site of both ER alpha and ER beta for all THC derivatives. Therefore, binding to this second site is probably not the reason the THC derivatives are agonists on ER alpha and antagonists on ER beta. Most likely, the smaller steroid binding site of ER beta compared to ER alpha and/or the apparent larger flexibility of helix 12 of ER beta make ER beta more readily adopt an antagonist conformation.

Rapid discovery of a novel series of Abl kinase inhibitors by application of an integrated microfluidic synthesis and screening platform.

Drug discovery faces economic and scientific imperatives to deliver lead molecules rapidly and efficiently. Using traditional paradigms the molecular design, synthesis, and screening loops enforce a significant time delay leading to inefficient use of data in the iterative molecular design process. Here, we report the application of a flow technology platform integrating the key elements of structure-activity relationship (SAR) generation to the discovery of novel Abl kinase inhibitors. The platform utilizes flow chemistry for rapid in-line synthesis, automated purification, and analysis coupled with bioassay. The combination of activity prediction using Random-Forest regression with chemical space sampling algorithms allows the construction of an activity model that refines itself after every iteration of synthesis and biological result. Within just 21 compounds, the automated process identified a novel template and hinge binding motif with pIC50 > 8 against Abl kinase--both wild type and clinically relevant mutants. Integrated microfluidic synthesis and screening coupled with machine learning design have the potential to greatly reduce the time and cost of drug discovery within the hit-to-lead and lead optimization phases.

TAK1 inhibition in the DFG-out conformation.

The first example of an inhibitor of the kinase TAK1 that binds in the DFG-out conformation is disclosed. These preliminary studies used kinase-targeted screening and structure-based drug design to create a molecule with dual pharmacological inhibition of p38 and TAK1 that demonstrated significant activity in a cell-based, anti-inflammatory assay.

Predicting cytotoxicity from heterogeneous data sources with Bayesian learning.

BACKGROUND: We collected data from over 80 different cytotoxicity assays from Pfizer in-house work as well as from public sources and investigated the feasibility of using these datasets, which come from a variety of assay formats (having for instance different measured endpoints, incubation times and cell types) to derive a general cytotoxicity model. Our main aim was to derive a computational model based on this data that can highlight potentially cytotoxic series early in the drug discovery process. RESULTS: We developed Bayesian models for each assay using Scitegic FCFP_6 fingerprints together with the default physical property descriptors. Pairs of assays that are mutually predictive were identified by calculating the ROC score of the model derived from one predicting the experimental outcome of the other, and vice versa. The prediction pairs were visualised in a network where nodes are assays and edges are drawn for ROC scores >0.60 in both directions. We observed that, if assay pairs (A, B) and (B, C) were mutually predictive, this was often not the case for the pair (A, C). The results from 48 assays connected to each other were merged in one training set of 145590 compounds and a general cytotoxicity model was derived. The model has been cross-validated as well as being validated with a set of 89 FDA approved drug compounds. CONCLUSIONS: We have generated a predictive model for general cytotoxicity which could speed up the drug discovery process in multiple ways. Firstly, this analysis has shown that the outcomes of different assay formats can be mutually predictive, thus removing the need to submit a potentially toxic compound to multiple assays. Furthermore, this analysis enables selection of (a) the easiest-to-run assay as corporate standard, or (b) the most descriptive panel of assays by including assays whose outcomes are not mutually predictive. The model is no replacement for a cytotoxicity assay but opens the opportunity to be more selective about which compounds are to be submitted to it. On a more mundane level, having data from more than 80 assays in one dataset answers, for the first time, the question - "what are the known cytotoxic compounds from the Pfizer compound collection?" Finally, having a predictive cytotoxicity model will assist the design of new compounds with a desired cytotoxicity profile, since comparison of the model output with data from an in vitro safety/toxicology assay suggests one is predictive of the other.

Designing compound subsets: comparison of random and rational approaches using statistical simulation.

Compound subsets, which may be screened where it is not feasible or desirable to screen all available compounds, may be designed using rational or random selection. Literature on the relative performance of random versus rational selection reports conflicting observations, possibly because some random subsets might be more representative than others and perform better than subsets designed by rational means, or vice versa. In order to address this likelihood, we simulated a large number of rationally designed subsets for evaluation against an equally large number of randomly generated subsets. We found that our rationally designed subsets give higher mean hit rates compared to those of the random ones. We also compared subsets comprising random plates with subsets of random compounds and found that, while the mean hit rate of both is the same, the former demonstrates more variation in the hit rate. The choice of compound file, rational subset method, and ratio of the subset size to the compound file size are key factors in the relative performance of random and rational selection, and statistical simulation is a viable way to identify the selection approach appropriate for a subset.