quantum molecular design of drugscomputer-aided drug design (cadd) is a specialized sub-discipline...

9
© 2015 Cloud Pharmaceuticals, Inc. Page1 WHITE PAPER Quantum Molecular Design of Drugs An In Silico Approach to Drug Discovery and Design in Novel Molecular Space Cloud Pharmaceuticals, Inc. 6 David Drive Research Triangle Park, NC www.cloudpharmaceuticals.com 1.919.424.6894 Email: [email protected]

Upload: others

Post on 03-Aug-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e1

W H I T E P A P E R

Quantum Molecular Design of Drugs An In Silico Approach to Drug Discovery and

Design in Novel Molecular Space

Cloud Pharmaceuticals, Inc. 6 David Drive

Research Triangle Park, NC www.cloudpharmaceuticals.com

1.919.424.6894 Email: [email protected]

Page 2: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e2

Contents

Searching Virtual “Chemical Space” ............................................................................................................ 3

Exploiting QM/MM ..................................................................................................................................... 4

Jak 3 Case Study Accurately Predicting Ligand Binding ............................................................................... 5

Quantum Molecular Design Offering .......................................................................................................... 9

Partnering ................................................................................................................................................... 9

Page 3: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e3

Searching Virtual “Chemical Space”

Chemical space is vast, with an estimated

1065 stable molecules accessible with

molecular weights below 850. Designing

new drugs that bind to a specified

protein target requires finding the best

molecule in this vast chemical space.

Exploration of this space by direct

enumeration and evaluation is

prohibitively costly. Cost effective

searching requires employing

optimization techniques.

Our novel "Quantum Molecular Design"

method can search large chemical space

much more efficiently. Quantum

Molecular Design uses "reverse

engineering" methods to solve the

problem of going from a set of desired

properties back to realistic chemical

structures and material morphologies

that may have these properties. The

specific implementation used by Cloud

Pharmaceuticals is based on the

pioneering work developed in Duke University, as shown in the figure taken from Wang et al, J. Am.

Chem. Soc. 2006.

The Beratan and Yang groups at Duke University have developed the linear combination of atomic

potentials (LCAP) approach, using molecular characteristics as a function of parameters that define the

contribution of a specific chemical group at a particular chemical site in a molecule. This method enables

the construction of a potentially enormous “virtual library” of chemical structures at a cost far below the

factorial cost of individual structure evaluation. The LCAP method has several advantages, such as

multiple search methods and ease of parallelization, and has been published and experimentally tested.

Cloud Pharmaceuticals has improved and enhanced the Quantum Molecular Design algorithm to allow

versatile and multiple chemical groups and has added a novel implementation that is based on an

integer programming method. This has allowed us to implement and reap the benefits of the algorithm

in previously unused areas, especially computational drug design.

Page 4: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e4

Exploiting QM/MM

The process of drug discovery

involves the identification of

molecular candidates, synthesis,

characterization, screening, and

assays for therapeutic efficacy. It

is generally recognized that drug

discovery and development are

very time and resource

consuming processes.

Computer-Aided Drug Design

(CADD) is a specialized sub-

discipline of rational drug design

that uses computational

methods to simulate drug-

receptor interactions and can

save time and money. Virtual

high throughput in silico

screening of ligand binding can significantly reduce the time required for lead discovery and lead

optimization. One of the most common tools of in silico binding analysis is the use of docking algorithms

to rapidly predict relative binding affinities of a large number of ligands for a given protein. If there is a

“hit” with a particular ligand, it can be extracted from the database for further testing. This molecule can

then go through ADMET (Adsorption, Distribution, Metabolism, Elimination and Toxicity) evaluation, as

well as synthesis, biological activity and refinement in order to generate a drug.

However, there is a major problem with how the drug design industry uses docking tools. While docking

tools are cheap in computer time and allow fast scans of large libraries, the accuracy of calculating the

binding strength (of ligands to therapeutic targets such as protein) is very poor and the results are not

predictive of the experimental data.

There are much more accurate methods than those currently used by the industry, but the cost (in

computer time) of such exhaustive calculations is very expensive, especially if one has to scan a very

large library of molecules. QM/MM methods are multi-scale/multi-resolution computational methods to

calculate ligand binding. Using a combination of quantum chemistry (QM) tools to characterize the

ligand, and molecular mechanics (MM) tools to describe the protein and solvent we obtain a deeper

understanding of protein-ligand interactions, which is also more accurate. The method includes

flexibility of the both the protein and ligand, as well as explicit water description.

Cloud Pharmaceuticals enables the efficient use of QM/MM by using it in tandem with the highly

efficient Quantum Molecular Design search algorithm.

Page 5: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e5

Jak 3 Case Study Accurately Predicting Ligand Binding

The successful treatment of diseases is highly dependent on the availability of effective medication,

often consisting of small molecules. Calculating the properties of all molecules in a large chemical library

can be time consuming and cost prohibitive. Cloud Pharmaceuticals solves this problem by using the

Quantum Molecular Design search algorithm to scan large libraries to find the strongest inhibitors of a

specific biological target and then calculate their binding strength using high accuracy QM/MM

calculations. We applied this methodology to the Janus family of kinases in order to discover novel

inhibitors of the JAK3 enzyme.

The Janus family of kinases includes JAK1, JAK2, JAK3, and TYK2. The JAK family is an active target for the

development of drugs for Rheumatoid Arthritis, immunosuppression and inflammation due to their

interaction with cytokine receptors. Cytokine receptors are instrumental in modulating the immune

system and are responsible for balancing humoral and cell-based immune responses and regulating the

maturation, growth and responsiveness of vital immune cell populations. However, these receptors

have no enzymatic activity and rely completely on the JAK enzymes to initiate signaling. Inhibitors

targeting JAK3 that do not target other members of the family, can produce more focused results with

fewer side effects.

JAK3 is involved in signaling IL-2 (T cell development), IL-4 (Th2 cell differentiation), IL-7 (thymocyte

development), IL-9 (hematopoietic cells), IL-15 (NK cell development), and IL-21 (immunoglobulin class

switching). JAK1 is involved in signaling all of these except IL-21 and also is part of the signaling

mechanism for IL-6, IL-10, IL-11, LIF, OSM, CT-1, CTNF, NNT-1, Leptin, and both type 1 and type2

interferon. JAK2 has been associated with the signaling of IL-3, IL-5, IL-6, and interferon as well as single

chain receptors (e.g. Epo-R, Tpo-R, GH-R, PRL-R). TYK2 is implicated in the signaling of interferon, IL-6, IL-

10, IL-11, IL-12, IL-27, IL-31, OSM, ciliary neurotrophic factor, cardiotrophin 1, cardiotrophin-like

cytokine, and LIF. Since JAK3’s signaling is limited to interleukins, targeting JAK3 allows for creating

drugs that potentiate interleukins without the unnecessary side effects causes by interruption of these

other receptor types.

Based on methodology developed at Duke University, Quantum Molecular Design is an innovative new

method that transforms the discrete chemical space to a continuous one, allows efficient searching of

that space, and cuts down the number of calculations required to locate promising lead structures in the

library. Quantum Molecular Design uses "reverse engineering" methods to solve the problem of going

from a set of desired properties back to realistic chemical structures and material morphologies that

may have these properties. This methodology also has the advantage of parallelizability. Computational

costs scale favorably with system size allowing for the use of highly parallel, very efficient computation

machines. This permits the use of the accurate, but computationally costly, quantum

mechanics/molecular mechanics (QM/MM) calculations resulting in very high accuracy binding affinity

assays of a ligand to a protein.

Page 6: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e6

Predicting binding affinities between a protein and a ligand is critical in order to rationally design new

drugs. Ligand binding is dominated by two terms: energy and entropy. In order to calculate the energy

and entropy that occurs during binding, the correct binding mode has to be known. The current

available solutions are divided broadly into two kinds:

Docking based methods (which are extremely fast, but not accurate, due to limitation of scoring

functions)

Free energy calculation methods, usually with thermodynamic integration (highly accurate, but

very time consuming)

Parameterization Process of Cloud Pharmaceuticals

Cloud Pharmaceuticals’ methodology gets the correct binding mode and binding energy (entropy) using

different ligand geometries in QM/MM energy calculations. Then we evaluate free energy (entropy)

contributions with surface area and solvation terms. Using multiple linear regressions, we calculate the

contribution of each term (binding energy, surface area and free-energy of solvation) for the best

prediction of ligand binding (for example, measured IC50). This equation can now be used to predict

what will be the binding strength of other ligands.

Page 7: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e7

For each target protein, X-ray structure is validated and corrected and then stripped of its X-ray ligand

and placed in a water box. After solvation, the waters and ions are equilibrated. Two ligand sets are

prepared for parameterization, a training set and a test set. The ligand binding sets are chosen based on

criteria of similarity to X-ray ligands, shared common core scaffold, diversity between the two data sets,

and the size of the data set. The conformers for each ligand in the data sets are generated, the solvation

term of the binding set is calculated, the solvent accessibility term is generated, and the QM/MM energy

term is calculated. These steps are performed for all of the conformers of all of the ligand in the two

data sets, which for Jak3 resulted in 3080 QM/MM parameterization runs, each one over 12 hours. A

linear regression is performed to get the best coefficients for each of the binding energy terms to

reproduce the experimental IC50.

This figure shows that QM/MM calculations

can predict, with high accuracy, the binding of

a set of JAK3 inhibitors. The top panel shows

the QM/MM calculation for a training set of 14

inhibitors (taken from Wang et al. Bioorg. Med.

Chem. Lett. 18 (2008) p. 4907), while the

bottom panel shows the same algorithm used

for a test set (taken from Antczak et al. Bioorg.

Med. Chem. Lett. 19 (2009) p. 6872). Using the

training data set we were able to fit our results

with a correlation of 0.77. The determined

coefficients for our energy term then

reproduce the test set data with a correlation

of 0.75.

The parameterized binding equation is used in

the Quantum Molecular Design methodology.

For finding novel Jak3 inhibitors, two virtual

libraries of compounds were built, a smaller

library built from a previously published

scaffold containing mostly unpublished

compounds and a larger library using a novel

scaffold, keeping the major ligand-protein

interactions from multiple available x-rays

structures. The smaller library can generate

approximately 1 million compounds by

modifying six different functional groups, whereas the larger library has seven different functional

groups and can generate approximately 36 million compounds.

Page 8: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e8

The libraries are then used in the Quantum Molecular Design algorithm, starting from multiple random

places in the library chemical space. The algorithm then goes through the library, improving the

optimized property. QM/MM calculations are performed on all of the conformers from the ligands

chosen by the algorithm. The parameterized binding equation is used to get a binding score that will

determine the direction along the property surface. In total, five runs with different initial starting

ligands were performed using the libraries. The top ligands with the strongest binding score were

chosen. After the top ligands are chosen they are run through a series of filters to determine drug

synthesizability, ADMET properties and intellectual property filters.

We also tested the same model on sets of JAK1, JAK2, and TYK2 inhibitors. These result were then used

to isolate molecules that precisely target JAK3 avoiding interaction with the rest of the JAK family of

enzymes. These ligands are available for review after signing NDA.

To summarize, we have successfully applied QM/MM computational methods to accurately predict the

measured strength of binding of a ligand to a protein, thus enabling the use of computers to identify

targeted inhibitors for JAK3, as well as numerous other pharmaceutical targets. Computational drug

design methods enable researchers to reduce the time and costs of the drug discovery process by

predicting experimental data in silico. Quantum Mechanics/Molecular Mechanics calculations are not

normally used in drug design because such models are computationally extensive, although (as we have

shown here) these methods offer much better accuracy than more commonly known computational

methods. We have reduced the computation burden by combining QM/MM with Quantum Molecular

Design (the search algorithm) tailored to high performance computation providing a computational

speed and cost advantage over other methods of drug discovery.

Drug Discovery Search Engine and Database

Drug Target

Protein Database Binding Database

Quantum Molecular Design

Process Structuresand Binding Data

Page 9: Quantum Molecular Design of DrugsComputer-Aided Drug Design (CADD) is a specialized sub-discipline of rational drug design that uses computational methods to simulate drug-receptor

© 2015 Cloud Pharmaceuticals, Inc.

Pag

e9

Quantum Molecular Design Offering

Cloud Pharmaceuticals offers its Quantum Molecular Design process as a service via Microsoft Azure,

Amazon EC2 and private cloud implementations. In a typical engagement, the customer provides a

target and its X-ray structure, and Cloud Pharmaceuticals analyzes the target using Quantum Molecular

Design, along with a number of “bioinformatics filters” to eliminate toxic leads and/or leads with poor

manufacturing properties. An analysis begins with the design of a scaffold for a small molecule or a

peptide, based on client choice. A calibration of the model is made validated by published or known

data or assays conducted by our experimental partners. The depth of search of molecular space is

determined by the customer’s budget. Upon completion of a customer engagement, typically a six

month effort for a single target, Cloud Pharmaceuticals provides a small, highly focused library of novel

lead compounds or peptide drug candidates for each target.

Quantum Molecular Design Workflow

Partnering

Cloud Pharmaceuticals partners with other biotechnology firms, pharmaceutical companies, medical

research institutions, and government laboratories to further develop leads in our pipeline. We have

applied Quantum Molecular Design to cancer, inflammation, autoimmune diseases, CNS indications and

rare diseases. For further information, contact our business development team, or write to

[email protected].