BIreactive: A Machine-Learning Model to Estimate Covalent Warhead Reactivity.

In the past decade, the pharmaceutical industry has paid closer attention to covalent drugs. Differently from standard noncovalent drugs, these compounds can exhibit peculiar properties, such as higher potency or longer duration of target inhibition with a potentially lower dosage. These properties are mainly driven by the reactive functional group present in the compound, the so-called warhead that forms a covalent bond with a specific nucleophilic amino-acid on the target. In this work, we report the possibility to combine ab initio activation energies with machine-learning to estimate covalent compound intrinsic reactivity. The idea behind this approach is to have a precise estimation of the transition state barriers, and thus of the compound reactivity, but with the speed of a machine-learning algorithm. We call this method "BIreactive". Here, we demonstrate this approach on acrylamides and 2-chloroacetamides, two warhead classes that possess different reaction mechanisms. In combination with our recently implemented truncation algorithm, we also demonstrate the possibility to use BIreactive not only for fragments but also for lead-like molecules. The generic nature of this approach allows also the extension to several other warheads. The combination of these factors makes BIreactive a valuable tool for the covalent drug discovery process in a pharmaceutical context.

Accurate and Rigorous Prediction of the Changes in Protein Free Energies in a Large-Scale Mutation Scan.

The prediction of mutation-induced free-energy changes in protein thermostability or protein-protein binding is of particular interest in the fields of protein design, biotechnology, and bioengineering. Herein, we achieve remarkable accuracy in a scan of 762 mutations estimating changes in protein thermostability based on the first principles of statistical mechanics. The remaining error in the free-energy estimates appears to be due to three sources in approximately equal parts, namely sampling, force-field inaccuracies, and experimental uncertainty. We propose a consensus force-field approach, which, together with an increased sampling time, leads to a free-energy prediction accuracy that matches those reached in experiments. This versatile approach enables accurate free-energy estimates for diverse proteins, including the prediction of changes in the melting temperature of the membrane protein neurotensin receptor 1.

pmx: Automated protein structure and topology generation for alchemical perturbations.

Computational protein design requires methods to accurately estimate free energy changes in protein stability or binding upon an amino acid mutation. From the different approaches available, molecular dynamics-based alchemical free energy calculations are unique in their accuracy and solid theoretical basis. The challenge in using these methods lies in the need to generate hybrid structures and topologies representing two physical states of a system. A custom made hybrid topology may prove useful for a particular mutation of interest, however, a high throughput mutation analysis calls for a more general approach. In this work, we present an automated procedure to generate hybrid structures and topologies for the amino acid mutations in all commonly used force fields. The described software is compatible with the Gromacs simulation package. The mutation libraries are readily supported for five force fields, namely Amber99SB, Amber99SB*-ILDN, OPLS-AA/L, Charmm22*, and Charmm36.

The impact of data integrity on decision making in early lead discovery.

Data driven decision making is a key element of today's pharmaceutical research, including early drug discovery. It comprises questions like which target to pursue, which chemical series to pursue, which compound to make next, or which compound to select for advanced profiling and promotion to pre-clinical development. In the following paper we will exemplify how data integrity, i.e. the context data is generated in and auxiliary information that is provided for individual result records, can influence decision making in early lead discovery programs. In addition we will describe some approaches which we pursue at Boehringer Ingelheim to reduce the risk for getting misguided.

Computational method to identify druggable binding sites that target protein-protein interactions.

Protein-protein interactions are implicated in the pathogenesis of many diseases and are therefore attractive but challenging targets for drug design. One of the challenges in development is the identification of potential druggable binding sites in protein interacting interfaces. Identification of interface surfaces can greatly aid rational drug design of small molecules inhibiting protein-protein interactions. In this work, starting from the structure of a free monomer, we have developed a ligand docking based method, called "FindBindSite" (FBS), to locate protein-protein interacting interface regions and potential druggable sites in this interface. FindBindSite utilizes the results from docking a small and diverse library of small molecules to the entire protein structure. By clustering regions with the highest docked ligand density from FBS, we have shown that these high ligand density regions strongly correlate with the known protein-protein interacting surfaces. We have further predicted potential druggable binding sites on the protein surface using FBS, with druggability being defined as the site with high density of ligands docked. FBS shows a hit rate of 71% with high confidence and 93% with lower confidence for the 41 proteins used for predicting druggable binding sites on the protein-protein interface. Mining the regions of lower ligand density that are contiguous with the high scoring high ligand density regions from FBS, we were able to map 70% of the protein-protein interacting surface in 24 out of 41 structures tested. We also observed that FBS has limited sensitivity to the size and nature of the small molecule library used for docking. The experimentally determined hotspot residues for each protein-protein complex cluster near the best scoring druggable binding sites identified by FBS. These results validate the ability of our technique to identify druggable sites within protein-protein interface regions that have the maximal possibility of interface disruption.

A designed conformational shift to control protein binding specificity.

In a conformational selection scenario, manipulating the populations of binding-competent states should be expected to affect protein binding. We demonstrate how in silico designed point mutations within the core of ubiquitin, remote from the binding interface, change the binding specificity by shifting the conformational equilibrium of the ground-state ensemble between open and closed substates that have a similar population in the wild-type protein. Binding affinities determined by NMR titration experiments agree with the predictions, thereby showing that, indeed, a shift in the conformational equilibrium enables us to alter ubiquitin's binding specificity and hence its function. Thus, we present a novel route towards designing specific binding by a conformational shift through exploiting the fact that conformational selection depends on the concentration of binding-competent substates.

Molecular structure of human GM-CSF in complex with a disease-associated anti-human GM-CSF autoantibody and its potential biological implications.

Polyclonal autoantibodies against human GM-CSF (granulocyte/macrophage colony-stimulating factor) are a hallmark of PAP (pulmonary alveolar proteinosis) and several other reported autoimmune diseases. MB007 is a high-affinity anti-(human GM-CSF) autoantibody isolated from a patient suffering from PAP which shows only modest neutralization of GM-CSF bioactivity. We describe the first crystal structure of a cytokine-directed human IgG1λ autoantibody-binding fragment (Fab) at 1.9 Å (1 Å=0.1 nm) resolution. Its CDR3-H substantially differs from all VH7 germline IgG1 structures reported previously. We derive a reliable model of the antigen-autoantibody complex by using NMR chemical shift perturbation data in combination with computational methods. Superposition of the modelled complex structure with the human GM-CSF-GM-CSF ternary receptor complex reveals only little overlap between receptor and Fab when bound to GM-CSF. Our model provides a structural basis for understanding the mode of action of the MB007 autoantibody.

Towards computational specificity screening of DNA-binding proteins.

DNA-binding proteins are key players in the regulation of gene expression and, hence, are essential for cell function. Chimeric proteins composed of DNA-binding domains and DNA modifying domains allow for precise genome manipulation. A key prerequisite is the specific recognition of a particular nucleotide sequence. Here, we quantitatively assess the binding affinity of DNA-binding proteins by molecular dynamics-based alchemical free energy simulations. A computational framework was developed to automatically set up in silico screening assays and estimate free energy differences using two independent procedures, based on equilibrium and non-equlibrium transformation pathways. The influence of simulation times on the accuracy of both procedures is presented. The binding specificity of a zinc-finger transcription factor to several sequences is calculated, and agreement with experimental data is shown. Finally we propose an in silico screening strategy aiming at the derivation of full specificity profiles for DNA-binding proteins.

ADMET Predictability at Boehringer Ingelheim: State-of-the-Art, and Do Bigger Datasets or Algorithms Make a Difference?

Computational methods assisting drug discovery and development are routine in the pharmaceutical industry. Digital recording of ADMET assays has provided a rich source of data for development of predictive models. Despite the accumulation of data and the public availability of advanced modeling algorithms, the utility of prediction in ADMET research is not clear. Here, we present a critical evaluation of the relationships between data volume, modeling algorithm, chemical representation and grouping, and temporal aspect (time sequence of assays) using an in-house ADMET database. We find no large difference in prediction algorithms nor any systemic and substantial gain from increasingly large datasets. Temporal-based data enlargement led to performance improvement in only in a limited number of assays, and with fractional improvement at best. Assays that are well-, intermediately-, or poorly-suited for ADMET predictions and reasons for such behavior are systematically identified, generating realistic expectations for areas in which computational models can be used to guide decision making in molecular design and development.

Evaluating the use of absolute binding free energy in the fragment optimisation process.

Key to the fragment optimisation process within drug design is the need to accurately capture the changes in affinity that are associated with a given set of chemical modifications. Due to the weakly binding nature of fragments, this has proven to be a challenging task, despite recent advancements in leveraging experimental and computational methods. In this work, we evaluate the use of Absolute Binding Free Energy (ABFE) calculations in guiding fragment optimisation decisions, retrospectively calculating binding free energies for 59 ligands across 4 fragment elaboration campaigns. We first demonstrate that ABFEs can be used to accurately rank fragment-sized binders with an overall Spearman's r of 0.89 and a Kendall τ of 0.67, although often deviating from experiment in absolute free energy values with an RMSE of 2.75 kcal/mol. We then also show that in several cases, retrospective fragment optimisation decisions can be supported by the ABFE calculations. Comparing against cheaper endpoint methods, namely Nwat-MM/GBSA, we find that ABFEs offer better ranking power and correlation metrics. Our results indicate that ABFE calculations can usefully guide fragment elaborations to maximise affinity.

An adapted consensus protein design strategy for identifying globally optimal biotherapeutics.

Biotherapeutic optimization, whether to improve general properties or to engineer specific attributes, is a time-consuming process with uncertain outcomes. Conversely, Consensus Protein Design has been shown to be a viable approach to enhance protein stability while retaining function. In adapting this method for a more limited number of protein sequences, we studied 21 consensus single-point variants from eight publicly available CD3 binding sequences with high similarity but diverse biophysical and pharmacological properties. All single-point consensus variants retained CD3 binding and performed similarly in cell-based functional assays. Using Ridge regression analysis, we identified the variants and sequence positions with overall beneficial effects on developability attributes of the CD3 binders. A second round of sequence generation that combined these substitutions into a single molecule yielded a unique CD3 binder with globally optimized developability attributes. In this first application to therapeutic antibodies, adapted Consensus Protein Design was found to be highly beneficial within lead optimization, conserving resources and minimizing iterations. Future implementations of this general strategy may help accelerate drug discovery and improve success rates in bringing novel biotherapeutics to market.

Conformational transitions upon ligand binding: holo-structure prediction from apo conformations.

Biological function of proteins is frequently associated with the formation of complexes with small-molecule ligands. Experimental structure determination of such complexes at atomic resolution, however, can be time-consuming and costly. Computational methods for structure prediction of protein/ligand complexes, particularly docking, are as yet restricted by their limited consideration of receptor flexibility, rendering them not applicable for predicting protein/ligand complexes if large conformational changes of the receptor upon ligand binding are involved. Accurate receptor models in the ligand-bound state (holo structures), however, are a prerequisite for successful structure-based drug design. Hence, if only an unbound (apo) structure is available distinct from the ligand-bound conformation, structure-based drug design is severely limited. We present a method to predict the structure of protein/ligand complexes based solely on the apo structure, the ligand and the radius of gyration of the holo structure. The method is applied to ten cases in which proteins undergo structural rearrangements of up to 7.1 A backbone RMSD upon ligand binding. In all cases, receptor models within 1.6 A backbone RMSD to the target were predicted and close-to-native ligand binding poses were obtained for 8 of 10 cases in the top-ranked complex models. A protocol is presented that is expected to enable structure modeling of protein/ligand complexes and structure-based drug design for cases where crystal structures of ligand-bound conformations are not available.

Protein thermostability calculations using alchemical free energy simulations.

Thermal stability of proteins is crucial for both biotechnological and therapeutic applications. Rational protein engineering therefore frequently aims at increasing thermal stability by introducing stabilizing mutations. The accurate prediction of the thermodynamic consequences caused by mutations, however, is highly challenging as thermal stability changes are caused by alterations in the free energy of folding. Growing computational power, however, increasingly allows us to use alchemical free energy simulations, such as free energy perturbation or thermodynamic integration, to calculate free energy differences with relatively high accuracy. In this article, we present an automated protocol for setting up alchemical free energy calculations for mutations of naturally occurring amino acids (except for proline) that allows an unprecedented, automated screening of large mutant libraries. To validate the developed protocol, we calculated thermodynamic stability differences for 109 mutations in the microbial Ribonuclease Barnase. The obtained quantitative agreement with experimental data illustrates the potential of the approach in protein engineering and design.

Ligand docking and binding site analysis with PyMOL and Autodock/Vina.

Docking of small molecule compounds into the binding site of a receptor and estimating the binding affinity of the complex is an important part of the structure-based drug design process. For a thorough understanding of the structural principles that determine the strength of a protein/ligand complex both, an accurate and fast docking protocol and the ability to visualize binding geometries and interactions are mandatory. Here we present an interface between the popular molecular graphics system PyMOL and the molecular docking suites Autodock and Vina and demonstrate how the combination of docking and visualization can aid structure-based drug design efforts.

The molecular mechanism of toxin-induced conformational changes in a potassium channel: relation to C-type inactivation.

Recently, a solid-state NMR study revealed that scorpion toxin binding leads to conformational changes in the selectivity filter of potassium channels. The exact nature of the conformational changes, however, remained elusive. We carried out all-atom molecular dynamics simulations that enabled us to cover the complete pathway of toxin approach and binding, and we validated our simulation results by using solid-state NMR data and electrophysiological measurements. Our structural model revealed a mechanism of cooperative toxin-induced conformational changes that accounts both for the signal changes observed in solid-state NMR and for the tight interaction between KcsA-Kv1.3 and Kaliotoxin. We show that this mechanism is structurally and functionally closely related to recovery from C-type inactivation. Furthermore, our simulations indicate heterogeneity in the binding modes of Kaliotoxin, which might serve to enhance its affinity for KcsA-Kv1.3 further by entropic stabilization.

Hybrid Screening Approach for Very Small Fragments: X-ray and Computational Screening on FKBP51.

Fragment-based drug discovery (FBDD) permits efficient sampling of the vast chemical space for hit identification. Libraries are screened biophysically and fragment:protein co-structures are determined by X-ray crystallography. In parallel, computational methods can derive pharmacophore models or screen virtual libraries. We screened 15 very small fragments (VSFs) (HA ≤ 11) computationally, using site identification by ligand competitive saturation (SILCS), and experimentally, by X-ray crystallography, to map potential interaction sites on the FKBP51 FK1 domain. We identified three hot spots and obtained 6 X-ray co-structures, giving a hit rate of 40%. SILCS FragMaps overlapped with X-ray structures. The compounds had millimolar affinities as determined by 15N HSQC NMR. VSFs identified the same interactions as known FK1 binder and provide new chemical starting points. We propose a hybrid screening strategy starting with SILCS, followed by a pharmacophore-derived X-ray screen and 15N HSQC NMR based KD determination to rapidly identify hits and their binding poses.

tCONCOORD-GUI: visually supported conformational sampling of bioactive molecules.

Conformational flexibility of bioactive molecules poses a major challenge to computational biology. tCONCOORD generates structure ensembles based on geometrical considerations and has been successfully applied to predict protein conformational flexibility and essential degrees of freedom. We have now developed a graphical user interface (GUI) for tCONCOORD, which substantially facilitates the simulation setup and provides valuable insights into the structure analysis and constraint definition process in tCONCOORD. Moreover, users can influence the constraint definition process by interactively turning interactions on and off, defining completely rigid or flexible regions, or by applying artifical constraints that cause a biased sampling of the conformational space. This interface offers a versatile environment not only for the setup and analysis of tCONCOORD simulations, but also for molecular modeling and structure analysis in general. Both tCONCOORD* and the tCONCOORD-GUI(dagger) are distributed freely.

Geometry-based sampling of conformational transitions in proteins.

The fast and accurate prediction of protein flexibility is one of the major challenges in protein science. Enzyme activity, signal transduction, and ligand binding are dynamic processes involving essential conformational changes ranging from small side chain fluctuations to reorientations of entire domains. In the present work, we describe a reimplementation of the CONCOORD approach, termed tCONCOORD, which allows a computationally efficient sampling of conformational transitions of a protein based on geometrical considerations. Moreover, it allows for the extraction of the essential degrees of freedom, which, in general, are the biologically relevant ones. The method rests on a reliable estimate of the stability of interactions observed in a starting structure, in particular those interactions that change during a conformational transition. Applications to adenylate kinase, calmodulin, aldose reductase, T4-lysozyme, staphylococcal nuclease, and ubiquitin show that experimentally known conformational transitions are faithfully predicted.

The thermodynamic influence of trapped water molecules on a protein-ligand interaction.

Water molecules doing time: Atomic-resolution crystal structures of the PPIase domain of cyclophilin G, alone and in complex with cyclosporin A, and together with MD simulations and calorimetry, reveal how trapped water molecules influence the thermodynamic profile of a protein-ligand interaction.

Fragment Binding Pose Predictions Using Unbiased Simulations and Markov-State Models.

Predicting the costructure of small-molecule ligands and their respective target proteins has been a long-standing problem in drug discovery. For weak binding compounds typically identified in fragment-based screening (FBS) campaigns, determination of the correct binding site and correct binding mode is usually done experimentally via X-ray crystallography. For many targets of pharmaceutical interest, however, establishing an X-ray system which allows for sufficient throughput to support a drug discovery project is not possible. In this case, exploration of fragment hits becomes a very laborious and consequently slow process with the generation of protein/ligand cocrystal structures as the bottleneck of the entire process. In this work, we introduce a computational method which is able to reliably predict binding sites and binding modes of fragment-like small molecules using solely the structure of the apoprotein and the ligand's chemical structure as input information. The method is based on molecular dynamics simulations and Markov-state models and can be run as a fully automated protocol requiring minimal human intervention. We describe the application of the method to a representative subset of different target classes and fragments from historical FBS efforts at Boehringer Ingelheim and discuss its potential integration into the overall fragment-based drug discovery workflow.