APR Sept/Oct 2023 - 65

« FORMULATION AND DEVELOPMENT
Machine learning, a subset of AI, encompasses software through
which a computer learns to predict similar or novel patterns in data
based on training on large datasets.13
In a drug discovery setting this
could mean, for example, correctly correlating " druggability " with
ligand-target interactions that were previously thought to be minor
or irrelevant. The goals of a machine learning system are to describe or
explain phenomena, predict what will occur when inputs are changed,
and to prescribe specific actions to achieve desired goals.
AI has the potential to supercharge and extract maximum actionable
information from both experimental and theoretical cloud-based
discovery applications, including those that are not natively
interrelated or interoperable. These include data from instrumental
methods (crystallography, nuclear magnetic resonance, etc.), that can
be used as inputs for docking, molecular dynamics, target- and ligandbased
pharmacophore characterization, quantitative structure-activity
relationships (QSAR), and similarity search.14
All these applications exist
in both " desktop " and cloud formats, but through the cloud users can
access them 24/7 without the need to purchase or maintain hardware
resources at their site.
Application modularity, which allows users to try or select from a host
of applications purchased and maintained by a service company,
becomes an even more compelling case for emerging discovery
applications, particularly those involving resource-intensive quantum
chemistry calculations.
While molecular mechanics (MM) is based on widely understood
classical mechanics,15
emerging QM approaches are not. QM takes
CADD and MM to a new level where chemical bonds are ignored.
Interacting species are instead described in terms of atomic nuclei and
electron clouds. QM methods provide the most accurate representation
of what is occurring in ligand-target interactions, including parameters
like vibrational frequencies, equilibrium molecular structure, dipole
moments, and reaction free energies, many of which aren't easily
accessible experimentally.16
QM is extremely resource-intensive and limited in terms of the
system complexity it can handle. To provide it with broader scope
and greater accessibility, developers have combined it with MM and
CADD to generate hybrid computational systems. Hybrid approaches,
for example QM-MM,17
QM virtual screening,18
and QM-QSAR,19
are
extremely complex and require dedicated staff for implementation
and maintenance, which is why they are best accessed through a
cloud-based service.
Cloud computing addresses a fundamental limitation of CADD. As
discovery organizations learned with LIMSs and ELNs, individual
site-based deployments are highly resource-intensive and limited to
whatever installed applications are available. Cloud-based services do
not require users to purchase, install, or maintain applications on site.
Security is arguably the primary concern for pharmaceutical companies
considering cloud computing. While cloud service providers have not
solved all security issues, the root causes of security breaches are fairly
well understood. According to a report by McKinsey Digital, " Almost
all breaches in the cloud stem from misconfiguration, rather than
from attacks that compromise the underlying cloud infrastructure. " 20
3.
4.
5.
6.
7.
8.
Conclusion
CADD has evolved from simple modeling, docking, and archiving
capabilities to a collection of applications involving very high dollar,
computational, and human resource overhead. Cloud computing is the
most efficient, cost-effective way to leverage advanced computational
applications, including hybrid quantum methods, without the need
to source, install, and maintain those applications onsite. The cloud
also provides a means of interrogating billion-compound virtual
libraries for novel chemical space. Rather than perceiving the cloud as
a potential source of data breaches, cloud users increasingly view the
security provided by cloud-based services as an improvement over inhouse
data security -- a bonus if you will.
References
1.
2.
Companies therefore " must adopt new security architectures and
processes to protect their cloud workloads. " Industrial consulting
group Gartner goes even further in stating that " 99% of cloud security
failures will be the customer's fault. " 21
In other words, actual security
breaches overwhelmingly arise from users' failure to assess risks
appropriately and to follow industry-standard cloud security practices.
Rather than viewing data security as a necessary " feature " of, or addon
to cloud discovery services, potential users need to consider
security itself as a service. From this perspective one can pose
similar questions around data safety as one would ask about the
cloud-resident discovery application: Do we have the resources to
implement a robust security plan ourselves, in house? Can a cloud
service provider do a better job than we can? A typical discovery
organization will probably answer " no " and " yes, " respectively, which
essentially eliminates the security conundrum.
DiMasi JA, Feldman L, Seckler A, Wilson A. Trends in risks associated with new drug
development:
success
rates
for
investigational
drugs. Clin Pharmacol Ther. 2010
Mar;87(3):272-7. doi: 10.1038/clpt.2009.295. Epub 2010 Feb 3. PMID: 20130567.
DiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry:
New estimates of R&D costs.
jhealeco.2016.01.012. Epub 2016 Feb 12. PMID: 26928437.
Mohs RC, Greig NH. Drug discovery and development: Role of basic biological research.
Alzheimers Dement (N Y). 2017 Nov 11;3(4):651-657. doi: 10.1016/j.trci.2017.10.005.
PMID: 29255791; PMCID: PMC5725284.
Abdalslam. LIMS Systems Statistics, Trends And Facts 2023. Accessed May 10, 2023, at
https://abdalslam.com/lims-systems-statistics.
Frye, L., Bhat, S,; Akinsanya, K,; Abel, R. From computer-aided drug discovery to computerdriven
drug discovery. Drug Discov. Today Technol. 2021, 39, 111- 117, DOI: 10.1016/j.
ddtec.2021.08.001.
Sliwoski G, Kothiwale S, Meiler J, Lowe EW Jr. Computational methods in drug discovery.
Pharmacol Rev. 2013 Dec 31;66(1):334-95. doi: 10.1124/pr.112.007336. PMID: 24381236;
PMCID: PMC3880464.
Spjuth O, Frid J, Hellander A. (2021) The machine learning life cycle and the cloud:
implications for drug discovery, Expert Opinion on Drug Discovery, 16:9, 1071-1079, DOI:
10.1080/17460441.2021.1932812.
Grance PMMT. NIST Definition of Cloud Computing. Special Publication (NIST SP) - 800145;
2011.
www.americanpharmaceuticalreview.com |
| 65
J Health Econ. 2016 May;47:20-33. doi: 10.1016/j.
»

http://www.americanpharmaceuticalreview.com

APR Sept/Oct 2023

Table of Contents for the Digital Edition of APR Sept/Oct 2023