Structure-Based Drug Design and Ligand Modification
Molecular Surfaces and Maps / Ligand Interactions / Docking / Ligand Optimization / Ligand Selectivity / Protein Alignments and Superposition
The course covers MOE applications for interactive structure based design. Examples include active site visualization, protein-ligand contact analysis and ligand modification/optimization in the receptor pocket. Use of the docking module and its application to assess ligand flexibility will be discussed. A protocol for aligning and superposing protein complexes in the context of protein selectivity will be studied.
The course will introduce cheminformatics applications using the MOE software. The topics covered include MOE
database manipulation, importing/exporting data to various formats (SDF, SMILES), molecular descriptor calculation,
data analysis and visualization, QSAR and binary-QSAR modeling, fingerprint-based similarity searching, clustering,
diverse subset selection and consensus modeling.
Introduction to SVL SVL Concepts / Scripts for Moe/Batch / Test New Ideas/Hypotheses / Customize MOE / Programming Techniques / Window Toolkit / Descriptors and Fingerprints / Interactive Problem Solving
Introduction to SVL
SVL Concepts / Scripts for Moe/Batch / Test New Ideas/Hypotheses / Customize MOE / Programming Techniques / Window Toolkit / Descriptors and Fingerprints / Interactive Problem Solving
The course covers both an overview of SVL capabilities and the context of its use in MOE, along with essential of the SVL language.
PSILO: A Protein-Ligand Structural Information Repository for Searching and Managing Public and Private Data
Macromolecular repository / 3D query searching / pocket similarity / display electron density / central repository / Specialized protein databases
PSILO® is a database system that provides an easily accessible, consolidated repository for macromolecular and protein-ligand structural information. The course will describe the application of PSILO® for mining data from macromolecular and protein-ligand structures. Approaches to query generation and search result navigation in PSILO will be covered in detail; for example, 3D contact searching and protein pocket similarity searching.
The course covers methods for analyzing and optimizing peptide-protein interactions in the active site. Topics related to peptide-protein structure preparation, peptide sequence optimization using natural and non-natural acids and conformational analysis will be discussed. Peptide-protein docking and protein-ligand interactions to analyze contact points will be described. The course will also cover advanced conformational searching using distance restraints.
The course describes advanced SBDD workflows in drug discovery projects and encompasses a range of topics from pharmacophore query generation to protein-ligand interaction fingerprints. More specifically, the course will cover the application of pharmacophores in the context of protein-ligand docking, scaffold replacement and R-group screening. A method for querying a 3D project database will also be presented along with the generation and analysis of protein- ligand interaction fingerprints (PLIF).
Protein Engineering / Protein Properties / Developability / Hot Spot Analysis / Antibody Modeling / Humanization / Molecular Surfaces
The course covers approaches for structure-based antibody design and includes protein-protein interactions analysis, in silico protein engineering, affinity modeling and antibody homology modeling. The interaction of a co-crystallized antibody-antigen complex will be studied by generating and examining the molecular surfaces and visualizing protein-protein interactions in 3D and 2D. Antibody properties will be evaluated using specialized calculated protein property descriptors and analyzing protein patches. The application of protein engineering tools for homology modeling and conducting property optimization of antibodies in the context of developability will be studied. Antibody optimization examples will include identification of glycosylation sites and analysis of correlated pairs using a specialized antibody database. An approach for humanizing antibody homology models will be discussed. All the steps necessary for producing and assessing antibody homology models will be described.
Ligand-Based Drug Design and SAR Analysis MOEsaic / R-Group Profiles and Analysis / MMP Analysis / Similarity and Substructure Searching / Descriptor Calculations / Conformational Searching / Molecular Alignments / Pharmacophore Modeling and Searching
Ligand-Based Drug Design and SAR Analysis
MOEsaic / R-Group Profiles and Analysis / MMP Analysis / Similarity and Substructure Searching / Descriptor Calculations / Conformational Searching / Molecular Alignments / Pharmacophore Modeling and Searching
The course covers essential in silico methods needed for guiding drug discovery projects in the absence of a protein structure. Analysis of SAR through R-group profiling and matched molecular pairs (MMP) analysis using MOEsaic to determine relationships among a chemical series are examined. Molecular descriptor calculations and their application for determining property correlations are described. Molecular alignments and conformational analyses of a congeneric series are explored to assess the impact of ligand substituents. An approach for developing pharmacophore queries is discussed. Management and manipulation of MOE databases are also covered.
Sequence and structure alignments / Template selection / Building and refining homology models
The course covers protein alignments and homology modeling in MOE. Alignment examples include multiple sequence and structure alignments, focusing alignments on a structurally conserved core or active site residues, and visualization of alignment properties such as percentage residue identity and RMSD. The homology modeling section covers the complete sequence to structure workflow, including template searching, query alignment, homology model building and refinement, along with applications for assessing model quality.
Biologics: Protein Alignments, Modeling and Docking
Protein Alignments and Superposition / Loop and Linker Modeling / Homology Modeling / Protein- Protein Docking / Protein Solubility Analysis / 2D Hot Spot Mapping / PLIF / Biologics QSAR Modeling
The course covers methods for aligning protein sequences, superposing structures, homology modeling fusion proteins and conducting protein-protein docking. In particular, an approach for aligning and superposing multiple structures will be described for determining structural and surface protein variations in relation to protein property modulation. A method for grafting and refining antibody CDR loops as well as using a knowledge-based approach to scFv fusion protein modeling using the MOE linker application will be described. An approach to generate homology models of a murine antigen structure from a human template as well as protein-protein docking of an antibody to an antigen will be discussed. A QSAR model for predicting and analyzing protein/biologics solubility will be described.
The course covers the suite of MOE applications which can be applied to small-molecule virtual screening. Topics include the preparation of small molecule databases for virtual screening, filtering databases based on substructure matching and property values, building QSAR/QSPR models and fingerprint similarity models as database filters, pharmacophore query creation and searching, and small-molecule docking. These tools are used in conjunction to present a complete virtual screening workflow. The creation of de novo structures using the MOE Scaffold Replacement and MOE Medchem transformation applications is also covered.
Helping Victims, Reforming Perpetrators using Rational Design
Roy Vaz, Consultant, Institute for Neurodegenerative Diseases, UCSF
When Drug-Drug interactions occur, one drug is usually the victim and the other is the
perpetrator. In other cases, the drug could be a victim of high pharmacogenetic variability, such
as being metabolized by CYP2D6 or CYP2C19. The perpetrator compound could be involved in
either inhibition (sometimes also time dependent) or induction of clearance enzymes.
Examples of Ameliorating compound series when the lead compound is a victim or possibly a
perpetrator will be described.
An example of changing clearance pathways so that the pharmacogenetically variable CYP2D6
enzyme was not involved in the end, in the clearance of a kinase compound (victim) will be
An example of changing a series of compounds (perpetrator) involved in metabolite-based
inhibition (MBI), will be shown.
Lastly, the rational modification of compounds (perpetrator) initially involved in CYP3A
induction via the PXR nuclear receptor so that the final compounds did not induce CYP3A, will
At the heart of small molecule lead discovery is an iterative design-synthesis-test cycle to establish structure activity relationship (SAR). During the learning cycle, computational chemists traditionally played a supporting role in the decision making progress and are often limited by synthesis resources. In the era of automated robotic compound synthesis and advancement on de novo design algorithms, we will present our recent effort to empower computational chemists and medicinal chemists to better make their decisions to drive the discovery project forward.
StructureTracker: A Design Platform for Drug Discovery Projects Scott Rowland, Associate Scientific Fellow, Takeda
Free Energy Calculations with Thermodynamic Integration in MOE using AMBER
Paul Labute, President and CEO, Chemical Computing Group
Until the advent of GPU accelerated molecular dynamics, alchemical free energy simulations had not been routinely applied in early stage drug discovery. While this GPU acceleration made such calculations more feasible, the complexity of the protocols and uncertainty regarding optimal parameters are a lingering problem.
In the recently released MOE 2019 a streamlined interface was added to set up free alchemical energy calculations dynamics simulations based on the Thermodynamic Integration (TI) method in AMBER 18. The workflow includes structure preparation, ligand parameterization, simulation planning and analysis. Free energy calculations are known to be extremely sensitive and have a high failure rate. We present an optimal simulation planning methodology that minimizes the expected error for a given system based on the available simulation resources (number of GPUs, nodes, etc.). Another major source of error is the simulation instability due to an imbalance between the electrostatic and van der Waals forces. Optimization of the AMBER soft-core potential resulted in much higher simulation stability and hence lower curvature of the potential function. In conjunction with Fejér numerical integration the evaluation of the alchemical integral can be achieved with spectral accuracy. The methodology was validated through the calculation of relative binding energies of 34 ligands of the p38 MAP Kinase and compared to experimental data.
Free or Not-So-Free: Affinity Predictions for Protein-Ligand Interactions
Essam Metwally, Associate Principal Scientist, Merck
The ability to accurately calculate the relative free energy of binding can be a useful in predicting ligand potency. As such several techniques strive to calculate relative free energies of binding as a surrogate predictive measure of the strength of a protein-ligand binding interaction. We examined and benchmarked several of these methods in an attempt to arrive at best-practices for prospectively determining which methods could or should be used on a drug discovery project a priori. We evaluated a dataset consisting of 15 drug targets with over 250 small molecule inhibitors representing diverse therapeutic areas and the performance of methods from simple scoring to computationally intensive free energy perturbation. The learnings with regards to predictivity, computational efficiency, and implementation will be discussed.
Molecular Modeling in Designing Chimeric Protein Degraders
Yilin Meng, Senior Scientist, Pfizer
Chimeric protein degraders (ProDegs, PROTACs) exert their efficacy by simultaneously binding an E3 ligase and a protein target of interest. This necessary first step results in protein target polyubiquitination and subsequent proteasome-mediated degradation. Compound optimization typically involves identifying a target binder and an E3 ligase binder and connecting the two pieces by varying linkers. Potent and selective degradation of various proteins have been reported in the literature and demonstrated in vivo. Recent structural studies of ternary complexes attempt to provide a detailed understanding of the molecular mechanism of action. This presentation will introduce this novel modality, summarize the state of the art in the field, and identify challenges and opportunities for computational modeling in this space with some specific case studies.
In silico Modeling of PROTAC-Mediated Ternary Complexes for Predicting Protein Degradation
Michael Drummond, Scientific Applications Manager, Chemical Computing Group
Recently, bifunctional small molecules known as Proteolysis-Targeting Chimeras, or PROTACs, have been the focus of intense research. PROTACs have the potential to offer a new modality in drug discovery, as a PROTAC tags a targeted protein for degradation rather than inhibition. Despite the demonstration of numerous advantages, the eventual success of the PROTAC approach hinges upon, among other factors, the ability to rationally modify and eventually design new PROTAC molecules. In this work, we will discuss our suite of tools for predicting the structures of ternary complexes, which are at the heart of successful protein degradation. In addition to generating a large ensemble of possible structures, we propose metrics that have been developed based on available experimental knowledge to identify the structures that are likely to degrade. We demonstrate the utility of our methods in a number of scenarios, including across different targets and PROTAC molecules.
MM-based Dipole Moment Calculations in Drug Design
Matt Lee, Director, Molecular Modeling & Design, CHDI Foundation
Calculated binding enthalpies for structure-based drug discovery using molecular mechanics (MM) force fields rely on the Born-Oppenheimer approximation to calculate energies as a function of atomic coordinates for bonded and non-bonded energy terms. The latter is derived entirely from van der Waals & Coulombic electrostatic terms and does not account for dipole dipole forces. Recent publications have characterized the role of quantum mechanics (QM)-based dipole moments in interaction energies. This study qualitatively compares intermolecular interactions of pi-systems i) for MM- vs. QM-based dipole moments and ii) for MM-based dipole moments vs. MM non-bonded Coulombic energies.
Water: Essential for Life and by Extension Drug Discovery
Daniel McKay, Investigator, Novartis
It is well accepted that life as we know it requires liquid water, it is less well understood as to why this is the case on a fundamental level. This work examines water on this level and shows that water is not only required for life as we know it but for life in general. Further this examination implies most of the methods used to account for water’s role are often insufficient for use within biochemical modeling and hence drug design.
Understanding Allosteric Interactions in hMLKL Protein That Modulate Necroptosis and Its Inhibition
Govinda Bhisetti, Principal Investigator and Head of Computational Chemistry, Biogen
Mixed Lineage Kinase Domain-Like protein (MLKL) is a multi-domain protein with an N-terminal 4 helical bundle (4HB) domain and a pseudokinase domain (PsK) connected by brace helices. It plays a key role in necroptosis - a non-apoptotic, caspase free and kinase dependent programmed cell death. Phosphorylation of PsK domain of MLKL leads to oligomerization of 4HB domain which migrates to plasma membrane and causes its rupture. Necrosulfonamide (NSA) binds to the 4HB domain of MLKL and prevents necroptosis. To understand the molecular details of MLKL function and its inhibition, we have performed a molecular dynamic study on human MLKL protein in apo, phosphorylated (activated) and NSA-bound states for a total 3 μs simulation time. Our simulations show increased inter-domain flexibility, increased rigidification of the activation loop, and increased alpha helical content in the brace helix region revealing a form of monomeric hMLKL necessary for oligomerization upon phosphorylation as compared to the apo state. NSA binding disrupts this activated form by causing two main effects on hMLKL conformation: (1) locking of the relative orientation of 4HB and PsK domains by the formation of several new interactions and (2) preclusion of key 4HB residues to participate in cross-linking for oligomer formation. This new understanding of the effect of hMLKL conformations on physphorylation and NSA binding suggests new avenues for designing effective allosteric inhibitors of hMLKL.
Society has witnessed through both news/media and the internet, the dangers and challenges of microbial and antibiotic resistance to therapeutic drugs in the human health sector. Considerably less attention has been given to the very real problem of agrochemical resistance is as it pertains to agricultural pest and weed management. The development of herbicide resistance in weeds challenges the sustainability of global food security for future generations. Modern herbicide resistance management programs focus on new biological targets or alternative "poisons" to old biological targets in an effort to counter the development of resistance in field settings. The shortcoming of these applied research directives is that they often fail to address key fundamental questions: Can we create new agrochemical solutions or identify pre-existing solutions that are less prone to resistance? Can we algorithmically and methodologically anticipate and avoid target-site resistance (TSR) in the agrochemical industry?
With a focus on identifying chemistries less prone to encountering resistance, we have selected the PPO (protoporphyrinogen oxidase) class of herbicides for which numerous herbicide-resistant weed strains exist and for which multiple classes of chemistries are on the market. Using workflows and tools developed for structure-based drug design, we have combined homology modeling, computational resistance scanning, mutation analysis and catalytic competency criteria to identify and structurally model a putative collection of biologically viable PPOX2 SNPs (single-nucleotide polymorphisms). These computationally validated/selected SNPs were used in a multi-target / multi-ligand (candidate PPO inhibitor chemistries) virtual screening campaign. The poses for the PPO chemistries explored were comparatively analyzed for each SNP against wild-type based on (a) protein-ligand interaction fingerprints, (b) impact on binding score and (c) ligand-to-substrate volumetric overlap. Results of this work have been validated against known field mutant/chemistry green-house experiments. Conclusions drawn from the approach aim at determining or promoting chemical candidate selections with a lower predicted susceptibility to resistance. Future work may include validation through bio-screening against cloned protein targets (plant/weed) and development of resistant mutations in yeast/microbes for testing novel chemistries at the target site.
This work illustrates, to our knowledge, the first application of computational protein design algorithms to prospectively predict plausible resistance mutations in planta under herbicide pressure. The approach demonstrated here could serve the dual purpose of designing new agrochemical solutions with a lower susceptibility for TSR while enabling the re-purposing of old but available agrochemical solutions at the known site of resistance.
Simulation of Competition Experiments in Order to Predict Relative Binding Affinities of Drug-Like Molecules
Hakan Gunaydin, Principal Scientist, Relay Therapeutics
Accurate prediction of relative binding affinities in the context of a fast-paced medicinal chemistry project team settings continues to be a challenge. The timelines from the conception to the registration of the target molecules can be as short as 1 day in fast-paced project teams with robust chemistry. In these kind of settings, state-of-the-art accurate relative binding affinity prediction methods (e.g. free energy perturbation - FEP) is of limited use even after the investment in computation resources such as GPUs. These type of timelines forces the prediction of relative binding affinities to be done with faster and less accurate methods. Hence, there is a need to establish relative binding affinity prediction methods that can be parallelized in order to achieve wall clock times that are conducive to making predictions in line with the pace medicinal chemistry teams synthesize molecules. The computation of relative binding affinities of drug-like molecules in the grand canonical monte carlo ensemble will be disclosed in this presentation. Here, we show that prediction of relative binding affinities with accuracies that are comparable to the ones obtained with FEP are achievable. Due to the nature of these calculations, grand canonical monte carlo calculations are infinitely parallelizable and offer an opportunity to make real-time predictions for the proposed targets in fast-paced medicinal chemistry project settings.
A Selectivity Prediction Model Based on MD simulations
Araz Jakalian, Principal Scientist (CADD), Paraza Pharma Inc.
A common criterion for inhibitor-design is selectivity against isoforms of the same protein. Using standard docking protocols to model the selectivity of compounds did not produce plausible explanations for the observed selectivity-SAR. Consequently, a molecular dynamics (MD) simulation approach was implemented to potentially offer an explanation for the experimental selectivity differences. In order to avoid energy convergence issues and be able to generate trajectories aligned with medicinal chemistry project timelines, a highly-approximated protein-ligand MD protocol was created using MOE’s MD engine. For each molecule, two MD simulations were performed, in protein-1 and protein-2 respectively. Time-averaged descriptors were extracted from the MD trajectories and correlations were found with the experimental selectivity data. The time-averaged electrostatics interaction energy between the ligands and protein-2 showed a good correlation with the experimental data with a r2=0.74 and q2=0.63. Adding a second element to the model, i.e. the change in standard deviation of the VDW interaction energy between the ligand in protein-2 and in protein-1, improved both the r2 and q2 of the model to 0.88 and 0.81 respectively. The MD QSAR model was able to capture and explain small structural changes in the molecule that caused large selectivity effects. Although the MD model is robust and predictive, we also discuss its limitations, in particular the limited applicability domain and some of the approximations that were used to simplify the protein-ligand construct.
Structure-Based Design of Small-Molecule Macrocycles and Simple Metrics to Identify Opportunities for Macrocyclization
Maxwell Cummings, Senior Principal Research Scientist, Janssen R&D
Macrocycles are of interest for drug discovery, particularly for protein-protein systems where larger ligands may be necessary to achieve drug-like binding affinity. Macrocyclization may offer an avenue to improved pharmaceutical properties for such larger small-molecule drugs. In drug discovery, macrocycles are often described as a single chemotype despite showing great diversity in chemical structure. Most macrocycle drugs arose from natural product-based drug discovery efforts, but a few are purely synthetic. We are interested in the structure-based design of purely synthetic small-molecule macrocycles, an area of drug discovery that has not been heavily explored. We present simple metrics that facilitate the detection and prioritization of bound ligands that may be particularly suited to macrocyclization. Representative examples of such macrocyclic-like compounds from both the Cambridge Structural Database (CSD) and Protein Data Bank (PDB) will be presented for discussion.
Structure Based Optimization Of TYK2 Pseudokinases Inhibitors From A DNA Encoded Library
Ninad Prabhu, Investigator, Medicine Design, GlaxoSmithKline
The JAK-family kinases, including TYK2, are unique in that they possess an ATP-binding JH2 domain in addition to a functional JH1 domain. JH2 ligand are found to negatively regulate activation of JH1 kinase domain. Inhibition of the JH2 domain may offer a unique opportunity to modulate the TYK2 kinase activity with high kinome selectivity.
This talk will describe the structure-based optimization of a TYK2 pseudokinase inhibitor identified from a DNA encoded library. The binding mode of the hit molecule was confirmed by X-ray crystallography. Optimization of the initial hit led to the identification of a chemical series with exceptional kinome selectivity, inhibiting the pro-inflammatory cytokines IL-12 and IL-23 whilst sparing JAK2-mediated EPO signaling.
Computational and Experimental Determination of the Impact of Sugars on the Conformational Stability of mAbs
Alejandro D'Aquino-Ruiz, Research Scientist, Janssen R&D
Understanding the role of various excipients in formulations is an important area of research in the pharmaceutical industry. Theories based on preferential exclusion and binding have been partially successful in explaining the role of excipients commonly used in formulation development of biotherapeutics. The exclusion theory predicts correlation between ΔGtransfer and ΔGunfolding measured at the same temperature. However, because under these conditions, ΔGunfolding must be obtained by chemical denaturation, requiring molar concentrations of denaturants, the effect of excipients at high concentration can only be determined for highly soluble excipients. As thermal denaturation has the potential to accommodate higher concentrations of excipient, it would be useful to understand the relationship between the change in Tm and the ratio of excipient to protein. Previous calculations using exclusion theory provided a linear relationship between the ΔGtransfer of sugars and the molecular mass of proteins. Here, we describe a combination of computational and experimental efforts to correlate a change in ΔTm measured during thermal stress for four monoclonal antibodies and 24 sugar-like excipients added to the solution. A strong correlation was found between the concentration of hydroxyl groups using 24 sugar-like excipients and the measured change in ΔTm for all the mAbs tested.
Molecular Basis for Kinase Specificity of Rock Inhibitors
Rafael Depetris, Principal Scientist I, Kadmon
The talk will describe our in silico and X-Ray crystallography based approach aimed to understand the mechanism of action of our inhibitors for the Rho-associated Coiled Coil Containing Kinase. We used a combination of these tools to guide our structure based drug design approach and managed to get inhibitors of ROCK which show little or no activity towards other members of the AGC family of serine threonine kinases. The talk with remark the importance of the structural analysis as a basis for drug development programs.
Structure Based Design of Potent Selective Inhibitors of Protein Kinase D1 (PKD1)
JW Feng, Director of Discovery Data Science, Denali Therapeutics
We previously disclosed a series of type I ½ inhibitors of NF-κB inducing kinase (NIK). Inhibition of NIK by these compounds was found to be strongly dependent on the inclusion and absolute stereochemistry of a propargyl tertiary alcohol as it forms critical hydrogen bonds (H-bonds) with NIK. We report that inhibition of protein kinase D1 (PKD1) by this class of compounds is not dependent on H-bond interactions of this tertiary alcohol. This feature was leveraged in the design of highly selective inhibitors of PKD1 that no longer inhibit NIK. A structure-based hypothesis based on the position and flexibility of the α-C-helix of PKD1 vs. NIK is presented.
Modeling Protein Properties using pH-dependent Conformational Sampling
John Gunn, Senior Research Scientist, Chemical Computing Group
Proteins present particular challenges for property calculations due to their conformational flexibility in solution and the sensitivity of the structure to environmental parameters such as buffer strength and pH. We present a novel method for calculating thermodynamically averaged properties using a conformational ensemble which correctly takes into account the variability of both the structure and the charge state (protonation) of the protein.
The validity of this approach will be demonstrated using various benchmark calculations with experimental reference data, and additional applications will be shown with an emphasis toward modeling developability criteria for therapeutic antibodies.
Discovery of Peptidomimetic Antibody–Drug Conjugate Linkers with Enhanced Protease Specificity
BinQing Wei, Senior Scientist, Genentech
Antibody–drug conjugates (ADCs) have become an important therapeutic modality for oncology, with three approved by the FDA and over 60 others in clinical trials. Despite the progress, improvements in ADC therapeutic index are desired. Peptide-based ADC linkers that are cleaved by lysosomal proteases have shown sufficient stability in serum and effective payload-release in targeted cells. If the linker can be preferentially hydrolyzed by tumor-specific proteases, safety margin may improve. However, the use of peptide-based linkers limits our ability to modulate protease specificity. Here we report the structure-guided discovery of novel, nonpeptidic ADC linkers. We show that a cyclobutane-1,1-dicarboxamide-containing linker is hydrolyzed predominantly by cathepsin B while the valine–citrulline dipeptide linker is not. ADCs bearing the nonpeptidic linker are as efficacious and stable in vivo as those with the dipeptide linker. Our results strongly support the application of the peptidomimetic linker and present new opportunities for improving the selectivity of ADCs.
Monoclonal antibodies are an important modality of human therapeutics. Understanding antibody-antigen recognition is
crucial for successful design and optimization of antibodies. This talk will describe learning that we've derived from
crystal structures of antibodies, especially antibody-antigen co-crystal structures, and propose how it can be used to
guide antibody modeling.
Role of Biophysical Assays and Simple Descriptors in Predicting Antibody Pharmacokinetics
Boris Grinshpun, Postdoctoral Fellow, EMD Serono
The study of pharmacokinetics (PK) focuses on understanding what happens to a substance after it is introduced into a living organism, including the processes of absorption, distribution, metabolism, and removal from tissues. Characterizing drug PK is therefore crucial for evaluating the therapeutic potential of any drug. For very large therapeutic proteins like antibodies the PK can be influenced by a number of complex and interrelated biophysical properties including charged and hydrophobic patches, aggregation propensity, conformational stability, and antigen binding affinity. Understanding the role of these properties in influencing drug kinetics in vivo would allow for early selection of molecules with favorable PK and reduce the number of molecules that fail in late stage clinical trials due to poor therapeutic exposure or efficacy. One of the key limitations for such a study is the lack of publicly available clinical PK data, and the high levels of variability in existing data due to differences in trial design, including patient pools, dosing regimens, and trial size. In this study we have collected and curated clearance data from clinical trials for a set of sixty-six antibodies in Phase 2 and above. We further investigate the relationship between the linear human clearance and a collection of in silico measured properties as well as published in vitro biophysical developability data for this antibody set. Our findings highlight the difficulty of extracting useful antibody PK data, and the complex relationship between antibody clearance, structure, and biophysical properties. Finally, we wish to emphasize the need in the industry to produce larger and more reliable public PK datasets of antibodies to allow for a better understanding of the molecular properties that influence their therapeutic half-life.
Antibody Humanization: Caveats & Strategies for Success
Stanley Krystek, Research Fellow, Molecular Structure & Design, Bristol-Myers Squibb
Computational protein design methods combined with mutational strategies will be discussed for the humanization/murinization of therapeutic antibodies. As our data demonstrate, an optimized computational protein design approach can be used to efficiently generate functional humanized antibodies. The computational approach has been used to provide humanized templates for affinity maturation and biophysical property optimization. Our process includes a combination of mutagenesis strategies coupled with high throughput screening for mutant evaluation. The outcome is a series of high quality, full length, humanized antibodies for expression.