ARTIFICIAL INTELLIGENCE IN LIFE SCIENCE
Artificial intelligence (AI) is an umbrella word used to describe machines and computer programs that exhibit, to a certain extent, cognitive thinking and decision-making. Notable examples are neural networks which may be trained to handle typically human tasks, ranging from research in material science to documents translation, to playing boards games such as Go.
AI is one of the driving forces of the fourth industrial revolution and virtually every industry is currently exploring its potential. Life science companies are no exception: in 2019 there was a 88%increase in the amount of healthcare organisation implementing an AI strategy. AI has proven itself valuable, if not disrupting, in several life science applications:
The present computational power and the state of the art neural networks allow for handling and analysing complex and big datasets. Hence, perhaps not surprisingly, AI is revolutionising the Pharma industry by leveraging the huge amount of data generated by genome sequencing techniques or by digitalising patient records:
Neural networks are already used in the discovery or design of potential drug candidates. For example, AtomNet is a deep learning convolutional Neural Networks that predicts active molecules for targets with no previously known modulators. In a nutshell, in the training phase, a Neural Network (NN) is trained to accomplish its task, e.g., in the case of AtomNet, the prediction of the bioactivity of small molecules for drug discovery applications. In the discovery phase, the trained neural network is fed with and manipulates an input dataset to accomplish its task, ultimately leading to the development of a product, e.g. a pharmaceutical or a chemical product.
For example, the input data fed to AtomNet represent a protein ligand pair. In the training phase, AtomNet is trained by using protein ligand pairs with known affinities and learns to predict the affinity of ligand pairs. In the discovery phase, AtomNet is fed with a potential protein ligand pair and estimate its affinity. AtomNet allows for exploring a big amount of potential protein ligand pairs to search for new molecules for targets with previously unknown modulators.
Business in life science is costly and risky, as it relies heavily on innovation. For example, in the Pharma industry, a drug takes about ten year and 2.5 billion dollars to be discovered, approved and launched on the market and only one out of ten drugs entering clinical trials is approved. In the life science business, intellectual property (IP) protection is widely and successfully used to mitigate risks and protect the economic investments e.g. by protecting research methods and their outcome. For example, in the last five years, the number of patent applications filed with the EPO in the pharmaceutical and in the biotech field has constantly increased.
The raising impact of AI in life science R&D makes imperative to formulate an IP strategy for appropriately protecting the AI algorithms as well as the data processed and the products developed by using these algorithms. Companies, and, especially, Universities and research institutes shall also revise the timing of steps of their IP strategies and consider IP protection already at the very early stages of their R&D activities. For example, assessing the affinity of ligand pairs is fundamental research which precedes and is propaedeutic to the development of a drug. Traditionally, such fundamental research activities are made by human scientists and are not always considered for IP protection, as no therapeutical product is identified yet. If, however, these activities are carried out by a neural network, the possibility to protect the algorithm, the data processed by the algorithm and, possibly, the outcome of the algorithm should be seriously taken into account. For instance, the company developing Atomwise has filed several patent applications aiming at protecting the neural network AtomNet.
IP PROTECTION FOR ARTIFICIAL INTELLIGENCE IN LIFE SCIENCE
First of all, it is worth investigating whether AI is patentable at all. According to the European Patent Office (EPO), if an invention involves AI contributing to the production of a technical effect that serves a technical purpose, the assessment of the invention’s patentability shall take AI into account. In practice, if the invention involves a Neural Network (NN) that accomplishes a task, the EPO will assess whether the accomplished task is technical. If this is the case, the NN is taken into account in the assessment of the patentability of the invention.
Where a Neural Network serves a technical purpose, the generation of the training data set and the steps of the training phase may also contribute to the technical character of the invention if they support achieving that technical purpose. The 2021 version of the Guidelines for the examination in the EPO lists several tasks that are considered to be technical:
Moreover, the primary Examiner of the (granted) patent application 15789480.9 considers the purpose to „predict binding affinity of one molecule to one target protein“ to be technical in nature.
Particular care shall be taken when drafting a patent application directed to an AI algorithm, as the patent application will be published and become part of the prior art. An inappropriately drafted patent application may be rejected or may hinder the patentability of the products obtained by using the AI algorithm described in the patent application.
Filing a patent application is not the only tool in the IP protection toolbox. In particular, depending on their business model and market, companies may decide to protect AI by trade secret. In a nutshell, a trade secret is information that (i) is not generally known, (ii) has commercial value and (iii) is handled by its holder to be kept secret. EU member countries protect trade secrets according to the Directive (EU) 2016/943 of June, 8 2016. In general, a trade secret shall be known to as few people as possible and, ideally, only to those who need to know it to do their job. Trade secrets‘ disseminations by former employees moved to competitors are not unusual and, typically, difficult to prevent and to prove.
Patent protection and trade secret are very different, if not antithetical, forms of IP:
Each of these forms of IP has its own advantages and drawbacks, that companies shall take into account when choosing the form of IP that fits best to their IP strategy and the invention to be protected.
It is worth noticing that software and, in particular, AI software is also protected by copyright. However, rather than to the idea underlying the software, this protection is directed to the actual implementation of this idea in the software and, hence is of limited use.
IP PROTECTION FOR DATA
According to the British mathematician and entrepreneur Clive Humby, „data is the new oil“. Like oil, data are valuable and need to be refined to become usable. Extracting and refining data are expensive activities that shall be protected. Several forms of IP may be used to achieve this protection:
Some companies restrict the access to their data by using trade secret, contract law and, obviously, technical protection means. In particular, they grant access to their data subject to license or transfer agreements. Data exclusivity allows for preventing some uses of clinical trial data generated for approval of drugs or chemical products.
The use of data to train AI algorithms may lead to copyright infringement and/or may infringe the sui-generis database right. Some legislations, however, provide for copyright exceptions, e.g. to foster scientific research. In 2021, EU member countries shall pass legislation meeting the requirements of the Directive (EU) 2019/790 on Copyright in the Digital Single Market. This Directive provides two copyright exceptions for text and data mining, which, in many cases, plays an important role in training machine learning algorithms.
On the one hand, companies shall take these exceptions into account when formulating their IP strategy and, for instance, expressly reserve their rights on their online content. On the other hand, some companies may profit from these exceptions to use proprietary data access provided by the Directive (EU) 2019/790.
It is worth noticing that, in many cases, data owned by Biotech companies may comprise personal information about individuals, e.g. patients, which have legal rights in these data, even when they do not own them. In Europe, data containing personal information shall be handled according to the General Data Protection Regulation (GDPR), a regulation in EU law aiming at giving individuals control on their personal information and at harmonising the regulations of the EU member countries.
The EPO provides for patent protection of computer-implemented data structures or formats embodied on a medium, provided that they meet the patentability requirements. if an invention involves a data structure or format contributing to the production of a technical effect that serves a technical purpose, the assessment of the invention’s patentability shall take this structure or format into account. This is the case, in particular, if the data structure or format controls the operation of the device processing the data. For example, according to the Guidelines for the examination in the EPO, an index structure used for searching a record in a database produces the technical effect of controlling the way the computer performs the search. A detailed discussion on the patentability of an exemplary data structure at the EPO and at the Chinese patent office can be found here.
CONCLUSIONS
When formulating their IP strategy, biotech companies, Universities and research institutes shall address the problem of appropriately protecting the AI algorithms as well as the data processed by and the outcome of these algorithms. Many forms of IP may be used to protect data and AI algorithms. These forms of IP have their own strengths and weaknesses that companies must take into account when choosing the form of IP that fits best to their IP strategy, to their business, to the market and to the to-be-protected invention.
This publication gives a general overview of the issues related to the IP protection of AI algorithms in life science. If you would like more information, please contact us at the following e-mail address: office@fleuchaus.com. We would be glad to put our experience and expertise in protecting computer implemented inventions and life science inventions at your disposal and provide you with advice tailored to your business needs.
References
- EPO’s Patent index 2020
- Niall McAlister and Roland Wiring, „AI in Life Sciences“
- Marta Duque Lizarralde, “A Guideline to Artificial Intelligence, Machine Learning and Intellectual Property”.
- Daniel Gervais, „Exploring the Interfaces Between Big Data and Intellectual Property Law“
- Raj Indupuri, „Clinical Research, Artificial Intelligence, and COVID-19“.
Authors
Dr. Andrea Fleuchaus
Deutsche und Europäische Patentanwältin, Europäische Marken- und Designanwältin
Edoardo Mirabella – European Patent Attorney