When it comes to protecting technical innovations there are two
broad strategies: keep it a trade secret or obtain a patent. It has
long been the case that computer-implemented technologies and
pharmaceutical technologies are technical fields where one or the
other approach (trade secrets in the case of computers, patents in
the case of pharmaceuticals) has been typically favoured. So, in
the brave new field of applying advanced computational systems to
the development of pharmaceuticals, how do we develop a strategy
which satisfies both approaches to protecting technical
To do so, we must look at the reasoning behind the preference
for each strategy and some of the problems faced by proprietors
when protecting their innovations in these technical fields.
There can be little doubt that sophisticated dynamic
computational systems, such as those dubbed “Artificial
Intelligence” (AI) and “Machine Learning”, are being
developed for use in all stages of drug design and development.
Advanced computational systems are being used to reduce the costs
associated with drug development, increase the number and variety
of candidates for further testing and improve testing and screening
of existing candidates, to name a few example applications.
Recent reports in the press would indicate that AI systems are
getting reasonably powerful at developing new and potent
biologically active compounds. For example, according to a report
by The Verge “AI suggested 40,000 new possible chemical
weapons in just six hours”.1 In this case,
researchers were exploring how artificial intelligence could be
used to develop biochemical weapons. Their findings were published
in the journal “Nature Machine
So, what sort of strategy should be adopted for a system that
can potentially identify 40,000 biologically active compounds in
six hours and what about the biologically active compounds
The rise of the application of AI and machine learning in the
fields of healthcare and drug development has inevitably led to a
corresponding rise in patent applications in this crossover field.
Therefore, questions regarding what and how to effectively protect
developments made in improving and adapting AI and machine learning
in various aspects of pharmaceutical development are being
frequently raised and discussed.
Where these computational systems assist inventors in a
meaningful way, it follows that there must be some technical and
patentable consideration of what parameters of the systems are
required for the effective use of the system. The development and
suitability of such systems for drug development must, many argue,
require technical and practical consideration.
Not only do the computational systems themselves represent
considerable value to applicants and inventors but also the ability
of such algorithms, software, or systems to develop new candidates
for drug therapies, screen existing candidates, provide in silico
models, or develop diagnostics (to name a few). In other words,
there is real value in the output of the system as well as the
system itself. However, claiming the output of the system based on
the virtue of the system itself runs headlong into well-established
principles of patent law.
Issues Relating to Patents
Patents and patent applications regarding AI and machine
learning face their own challenges. In 2018, the European Patent
Office (EPO) amended their Guidelines for Examination to include a
section on artificial intelligence and machine learning making
clear that such inventions would be considered as other
computer-implemented inventions and assessed under the EPO’s
well-established framework considering technical character,
technical contribution and technical effect of the subject matter
of the invention.
G-II, 3.3.1 of the 2022 edition of the EPO’s Guidelines for
Examination provides the following guidance on potential technical
application for AI and machine learning:
“Artificial intelligence and machine learning find
applications in various fields of technology. For example, the use
of a neural network in a heart monitoring apparatus for the purpose
of identifying irregular heartbeats makes a technical contribution.
The classification of digital images, videos, audio or speech
signals based on low- level features (e.g. edges or pixel
attributes for images) are further typical technical applications
of classification algorithms.”
Meanwhile, in 2020, the United States Patent and Trademark
Office (USPTO) released a report on “Public Views on
Artificial Intelligence and Intellectual Property Policy”,
which also indicated that “AI inventions should not be treated
any differently than other computer-implemented invention”.
Thus, the USPTO determines patent-eligibility of AI and machine
learning related inventions based on the existing test which
evaluates whether such inventions fall into the judicial
exceptions, such as being an abstract idea. Finding that an
invention has a practical application often helps to ensure that
the invention falls outside the judicial exceptions, and thus is
The patent system therefore appears to be appropriate for
protection of innovation relating to AI and/or machine learning
systems which are applied to the practical and technical problems
associated with pharmaceutical development.
It is worth noting that the utilisation of AI and machine
learning systems in pharmaceutical development, as with other
computer-implemented inventions, may have a stronger and more
straight forward case for falling outside the judicial exceptions
in the US than being found to provide a technical solution to a
technical problem as required at the EPO. Some systems may
therefore have a better chance before the USPTO than the EPO and as
such may require different patentability considerations for each
Claims to Products Resulting from an AI/ ML Process
Patents appear to be a suitable means for protecting the
underlying AI or machine learning system directed to solving a
practical or technical problem in developing or screening
pharmaceutical products, but what about the products
Both the USPTO and EPO require some structure or technical
content to be imparted on the product by the inventive process in
order for such product claims to be found patentable.
Guidelines for Examination at the EPO recite:
“The technical content of the invention lies not in the
process per se, but rather in the technical properties imparted to
the product by the process.” (F-IV, 4.12)
Whilst the Manual of Patent Examining Procedure at the USPTO
“Product-by-process claims are not limited to the
manipulations of the recited steps, only the structure implied by
the steps.” (Section 2113)
Therefore, claims directed to the products of the AI or machine
learning systems must have some structural and technical property
imparted on them by virtue of being produced by the AI or machine
learning system which renders the product novel and inventive in
its own right.
Application of Known AI/ML Process to Novel Purpose
A further consideration for the application of AI or machine
learning processes in the development of pharmaceuticals is the
application of such processes in which the process is known but the
application is novel.
At the EPO, the novel purpose of a known process would have to
apply such a known process in a non-obvious way and have for
example a new and surprising effect or would have to overcome
technical difficulties not resolvable by routine techniques.
AI/ML as a Routine Technique
AI or machine learning inventions directed to developing new
drugs or otherwise deriving their technical character from
undisclosed applications and effects run a high risk of facing
patentability issues in the EPO or AI or machine learning
inventions directed to developing new drugs or otherwise deriving
their technical character from undisclosed applications and effects
run a high risk of facing challenges in the EPO. The EPO would
consider whether the disclosure is sufficient that it was plausible
that such effects would arise or that such effects arise across the
entire scope of the claims.
The challenge for inventions of this nature will be to navigate
inventive step and sufficiency requirements. The applicant must try
to show that the skilled person would understand the undisclosed
and technical application well enough to implement the invention.
The applicant must also prove that the functional elements of the
system are sufficiently distinct from the prior art meaning the
skilled person would not find the system itself to be obvious over
the prior art.
In other words, could generic or known AI systems be considered
tools for routine experimentation like other known tools and
thereby render a novel and inventive application of such tools
If so, are there aspects of such systems that are not routine
and have a new and surprising effect? An applicant must consider
what elements of the utilised AI/machine learning system are known
or routine and what elements differ in their application
sufficiently to be suitable grounds of a patent application. Above
all else, such a distinction must be described in sufficient
technical detail and the new and inventive patentable aspects
delineated from that which is known or routine.
Trade Secrets and Data Access
There are valuable aspects of the innovation associated with
computerised systems implemented in the development of valuable end
products that may not be suitable for the patent system alone in
its current form.
Provisions for trade secrets provide an alternative and/or
supplementary mechanism to the patent system for protecting
innovations of a technical nature.
The protection afforded by trade secret provisions is automatic
for as long as the requirements of the relevant trade secret laws
are met e.g., keeping the relevant information secret.
However, technology that requires regulatory approval (and
therefore detailed disclosure), may not be suitable for complete
reliance on trade secret provisions. Pharmaceutical technology, for
example, cannot rely on trade secret provisions to the extent that
the technology needs to be disclosed to regulators for market
US and European Trade Secret Law
In the US, a trade secret is information that the proprietor has
taken reasonable steps to keep secret and that has commercial value
by virtue of it being a secret. However, as mentioned above,
protection only lasts for as long as these conditions are met.
Information would stop being a trade secret if it ceased to be a
secret or if the owner failed to maintain reasonable measures to
maintain its secrecy.3
Similar standards for trade secrets have been introduced into
European law by EU Directive (2016/943).
In accordance with the EU Directive 2016/943 on trade secrets,
all EU members at the time (including the UK) unified their
provisions for the protection of trade secrets and brought their
requirements largely in line with those in the US.
Namely, the requirements that a trade secret be information that
is (i) secret (ii) has commercial value because it is secret and
(iii) has been subject to reasonable steps under the circumstances
to keep it secret (see Article 2 of EU Directive 2016/943).
A key consideration, therefore, for highly valuable information
relating to technology for the development of pharmaceutical
products is whether the relevant information can be successfully
kept secret and what reasonable steps must be maintained.
Pharmaceutical innovations usually have an associated obligation
of disclosure to regulators and remain commercially significant
throughout the term of a patent and beyond, whilst
computer-implemented technologies can be made difficult to reverse
engineer (such as with SaaS models) and access can more easily be
restricted so the risk of inadvertent disclosure to the public can
be significantly reduced. Computer- implemented technologies are
also more often made redundant before the end of the term of a
Figure 1 below illustrates an example IP strategy for an AI or
Machine Learning technology with a pharmaceutical application:
In relation to taking reasonable steps for technology that is
suitable for being kept secret, examples of reasonable steps may
include: identifying and labelling secret information; informing
and agreeing confidentiality with those who have access; and
managing security risks such as encrypting the relevant
As with all matters of legal enforcement, the collection and
maintenance of evidence is key. In the case of trade secrets this
means data about the data or metadata should be collected and
It should also be noted that, according to a research article
from the University of Oxford, the more widely information is
shared within an organisation, the less time until eventual
disclosure.1 Table 4 of the article indicates that to keep
something secret for 20 years, no more than 628 people can have
knowledge of the secret.
In consideration of all of the above, a strategy including
patent and trade secret provisions across both Europe and the US is
most likely to be the best way to sufficiently protect valuable
developments of AI and machine learning systems in the field of
A significant consideration in such a strategy should balance
the practicalities of protecting secret information from
unauthorised disclosure over time and the suitability of
information for patent protection.
Where possible and when trade secret laws are being relied upon,
access to information should be limited on a need-to- know basis
and where there is a significant risk of disclosure or leaking of
information a patent strategy should be formulated.
3. For more on legal protections for US trade secrets
please see here: https://www.finnegan.com/a/web/316168/PUBLISHED-Bloomberg-
10.1371/journal.pone.0147905 [On the Viability of
Conspiratorial Beliefs – David Robert Grimes, University of
The content of this article is intended to provide a general
guide to the subject matter. Specialist advice should be sought
about your specific circumstances.