AI in Drug Discovery | IP considerations
Dr. Evangelo Damigos; PhD | Head of Digital Futures Research Desk
- Sustainable Growth and Tech Trends
Publication | Update: Sep 2020
According to CMS research, drug manufacturers already use AI to aid the discovery of compounds that could potentially lead to new drugs, or to discover new uses for known drugs.
AI has huge potential to disrupt this industry, where roughly 9 in 10 of every drug candidates fails to make it to market and where billions of dollars is invested for each drug that eventually comes to market.
According to Prescouter, the use of AI in the pharmaceutical industry is projected to bring in billions of dollars in funding in the near future, underlying the huge potential for growth in this specific sector. AI has the potential to transform the drug development process making it both more efficient and more effective, thus benefiting all parties involved — from the companies developing new drugs to the patients in desperate need of viable treatments.
There are many ways in which AI is used to aid drug development and discovery. The methods discussed below are merely a sample of the many ways of solving the problem of how to improve drug discovery.
BenevolentAI, a UK-based firm that recently announced a collaboration with AstraZeneca, aims to use AI at all stages of drug development, from discovery to late stage clinical development. This firm uses machine learning to mine and analyse biomedical information from clinical trials data and academic papers. The AI can then identify molecules that have failed clinical trials and predict whether those molecules would be more efficient in targeting other diseases. The predictive power of the AI can also be used to design new molecules. The collaboration with AstraZeneca is intended to discover new drugs for chronic kidney disease and idiopathic pulmonary fibrosis.
Atomwise is a company which approaches drug discovery from a very different perspective. It aims to find new molecules for diseases which have been hard to target, especially “undrugged” proteins, and has recently announced a partnership with Eli Lilly. Atomwise have applied methods for speech and image recognition to chemistry to predict the bioactivity of small molecules. Their technology is based on convolutional neural networks and extracts insights from millions of experimental affinity measurements and analysis of thousands of protein structures to predict the binding of small molecules to proteins.
Recursion Pharmaceuticals represents yet another radically different approach to drug development. They use high-throughput screening methods to generate hundreds of thousands of cellular images which their AI software analyses. The AI allows development of cellular models of diseases, which provide insights into toxicity, dose-response characteristics, and target discovery and prediction. Their entirely automated approach is more similar to traditional
drug screening methods, but is scaled up to allow thousands of compounds to be screened against hundreds of diseases cheaply and quickly.
AI has huge potential to transform and disrupt the drug discovery process resulting in a dramatic acceleration in the time taken to develop new drugs and enormous cost reductions for the development of new drugs. It is estimated that the pre-clinical stage accounts for 33% of the cost of developing a new drug.
The estimated cost for developing a new drug is US$ 2.6 billion. Nic Fleming authored an article published in Nature entitled "How Artificial Intelligence is changing drug discovery" that observed "Few people in the field doubt the need to do things differently."
Drug Discovery entails a big data challenge with a vast search space. Olğaç et al. authored research entitled "Cloud-Based High Throughput Virtual Screening in Novel Drug Discovery".
Image Source: Pharma.org The Biopharmaceutical Research and Development Process
An article in Bloomberg entitled "AI Drug Hunters Could Give Big Pharma a Run for Its Money" further emphasised the points made above noting "And science moves slowly: In the nearly 20 years since the human genome was sequenced, researchers have found treatments for a tiny fraction of the approximately 7,000 known rare diseases."
Further, there are approximately 20,000 genes that can malfunction in at least 100,000 ways, and millions of possible interactions between the resultant proteins. It’s impossible for drug hunters to probe all of those combinations by hand.
Artificial Intelligence could be used to scan millions of high-resolution cellular images—more than humans could ever process on their own—to identify therapies that could make diseased cells healthier in unexpected ways.
Recursion, a startup applying Machine Learning techniques to scan images and search compounds that may disrupt disease without harming healthy cells raised 1 million in its latest financing round at a valuation of 6 million, according to PitchBook.
At DLS, we have also worked with medical imaging with Deep Learning techniques applied for cell imaging and believe that computer vision in healthcare is set for continued growth.
Ingrid Torjsen "Drug development: the journey of a medicine from lab to shelf" summarised a part of the drug discovery process and noted that upon identification of a potential target, researchers embark upon a search for a compound or molecule that will act upon the given target. In the past researchers search for candidate drugs focussed on natural compounds for example extracted from plants or fungi, however, the recent trend has been for researchers to apply knowledge obtained from the studying genetics as well as proteins to develop new molecules via computers. The process results in up to 10,000 compounds being considered and reduced to a mere 10 to 20 that have a theoretical ability to intervene in the disease process.
IP considerations – Patents: The compounds that are discovered are an obvious target for patent protection for any company using AI in drug discovery. In this respect, drugs are a good opportunity to get patent protection on products discovered via AI. BenevolentAI, for example, have several patent applications in their name which all appear to be directed solely towards the products of their AI (i.e., compounds and uses thereof).
Companies may also decide to use patents to protect their AI. Atomwise and Recursion Pharmaceuticals appear to have patent applications directed towards their AI, whereas BenevolentAI does not appear to have any pending applications which disclose their AI.
Patents, by their nature, require public disclosure of a company’s methods or product. There is a possibility that by disclosing an AI for drug discovery would make any drugs resulting from that AI obvious, if for example it could be demonstrated that the AI would always arrive at the same result when given a problem and that a skilled person would inevitably arrive at that solution if he/she were given the problem and the AI. Anyone seeking to protect their AI should therefore take care that their disclosure of the AI does not render any drugs resulting from the AI obvious.
When considering patent protection, it is also important to remember that, in Europe, an invention must be directed towards a non-obvious technical solution of a technical problem. In this respect, all features which contribute to the technical character of the invention are taken into account. The EPO’s guidance is that AI-related innovations should be described and claimed as being developed for a specific implementation. However, a pending Atomwise EP application has been deemed to relate to non-technical matter because the claim is “not sufficiently limited to ensure that [the] technical purpose is actually served by the distinguishing features over the whole claim scope”. Applicants should therefore ensure the technical purpose is plausibly served across the whole scope of the claimed subject matter.
AI users may also consider using trade secrets to protect their AI. Trade secrets do not prevent third parties from independently arriving at the same solution, however, and anyone using trade secrets should also be aware of the dangers of third parties reverse-engineering their AI.
In summary, patent protection is available for AI technologies in the field of drug discovery and companies operating in this area should seek expert guidance at the earliest opportunity to discuss their options and any difficulties they may encounter.
Objectives and Study Scope
This study has assimilated knowledge and insight from business and subject-matter experts, and from a broad spectrum of market initiatives. Building on this research, the objectives of this market research report is to provide actionable intelligence on opportunities alongside the market size of various segments, as well as fact-based information on key factors influencing the market- growth drivers, industry-specific challenges and other critical issues in terms of detailed analysis and impact.
The report in its entirety provides a comprehensive overview of the current global condition, as well as notable opportunities and challenges.
The analysis reflects market size, latest trends, growth drivers, threats, opportunities, as well as key market segments. The study addresses market dynamics in several geographic segments along with market analysis for the current market environment and future scenario over the forecast period.
The report also segments the market into various categories based on the product, end user, application, type, and region.
The report also studies various growth drivers and restraints impacting the market, plus a comprehensive market and vendor landscape in addition to a SWOT analysis of the key players. This analysis also examines the competitive landscape within each market. Market factors are assessed by examining barriers to entry and market opportunities. Strategies adopted by key players including recent developments, new product launches, merger and acquisitions, and other insightful updates are provided.
Research Process & Methodology
We leverage extensive primary research, our contact database, knowledge of companies and industry relationships, patent and academic journal searches, and Institutes and University associate links to frame a strong visibility in the markets and technologies we cover.
We draw on available data sources and methods to profile developments. We use computerised data mining methods and analytical techniques, including cluster and regression modelling, to identify patterns from publicly available online information on enterprise web sites.
Historical, qualitative and quantitative information is obtained principally from confidential and proprietary sources, professional network, annual reports, investor relationship presentations, and expert interviews, about key factors, such as recent trends in industry performance and identify factors underlying those trends - drivers, restraints, opportunities, and challenges influencing the growth of the market, for both, the supply and demand sides.
In addition to our own desk research, various secondary sources, such as Hoovers, Dun & Bradstreet, Bloomberg BusinessWeek, Statista, are referred to identify key players in the industry, supply chain and market size, percentage shares, splits, and breakdowns into segments and subsegments with respect to individual growth trends, prospects, and contribution to the total market.
Research Portfolio Sources:
Global Business Reviews, Research Papers, Commentary & Strategy Reports
M&A and Risk Management | Regulation
The future outlook “forecast” is based on a set of statistical methods such as regression analysis, industry specific drivers as well as analyst evaluations, as well as analysis of the trends that influence economic outcomes and business decision making.
The Global Economic Model is covering the political environment, the macroeconomic environment, market opportunities, policy towards free enterprise and competition, policy towards foreign investment, foreign trade and exchange controls, taxes, financing, the labour market and infrastructure. We aim update our market forecast to include the latest market developments and trends.
Review of independent forecasts for the main macroeconomic variables by the following organizations provide a holistic overview of the range of alternative opinions:
As a result, the reported forecasts derive from different forecasters and may not represent the view of any one forecaster over the whole of the forecast period. These projections provide an indication of what is, in our view most likely to happen, not what it will definitely happen.
Short- and medium-term forecasts are based on a “demand-side” forecasting framework, under the assumption that supply adjusts to meet demand either directly through changes in output or through the depletion of inventories.
Long-term projections rely on a supply-side framework, in which output is determined by the availability of labour and capital equipment and the growth in productivity.
Long-term growth prospects, are impacted by factors including the workforce capabilities, the openness of the economy to trade, the legal framework, fiscal policy, the degree of government regulation.
Direct contribution to GDP
The method for calculating the direct contribution of an industry to GDP, is to measure its ‘gross value added’ (GVA); that is, to calculate the difference between the industry’s total pretax revenue and its total boughtin costs (costs excluding wages and salaries).
Forecasts of GDP growth: GDP = CN+IN+GS+NEX
GDP growth estimates take into account:
All relevant markets are quantified utilizing revenue figures for the forecast period. The Compound Annual Growth Rate (CAGR) within each segment is used to measure growth and to extrapolate data when figures are not publicly available.
Our market segments reflect major categories and subcategories of the global market, followed by an analysis of statistical data covering national spending and international trade relations and patterns. Market values reflect revenues paid by the final customer / end user to vendors and service providers either directly or through distribution channels, excluding VAT. Local currencies are converted to USD using the yearly average exchange rates of local currencies to the USD for the respective year as provided by the IMF World Economic Outlook Database.
Industry Life Cycle Market Phase
Market phase is determined using factors in the Industry Life Cycle model. The adapted market phase definitions are as follows:
The Global Economic Model
The Global Economic Model brings together macroeconomic and sectoral forecasts for quantifying the key relationships.
The model is a hybrid statistical model that uses macroeconomic variables and inter-industry linkages to forecast sectoral output. The model is used to forecast not just output, but prices, wages, employment and investment. The principal variables driving the industry model are the components of final demand, which directly or indirectly determine the demand facing each industry. However, other macroeconomic assumptions — in particular exchange rates, as well as world commodity prices — also enter into the equation, as well as other industry specific factors that have been or are expected to impact.
Forecasts of GDP growth per capita based on these factors can then be combined with demographic projections to give forecasts for overall GDP growth.
Wherever possible, publicly available data from ofﬁcial sources are used for the latest available year. Qualitative indicators are normalised (on the basis of: Normalised x = (x - Min(x)) / (Max(x) - Min(x)) where Min(x) and Max(x) are, the lowest and highest values for any given indicator respectively) and then aggregated across categories to enable an overall comparison. The normalised value is then transformed into a positive number on a scale of 0 to 100. The weighting assigned to each indicator can be changed to reﬂect different assumptions about their relative importance.
The principal explanatory variable in each industry’s output equation is the Total Demand variable, encompassing exogenous macroeconomic assumptions, consumer spending and investment, and intermediate demand for goods and services by sectors of the economy for use as inputs in the production of their own goods and services.
Elasticity measures the response of one economic variable to a change in another economic variable, whether the good or service is demanded as an input into a final product or whether it is the final product, and provides insight into the proportional impact of different economic actions and policy decisions.
Demand elasticities measure the change in the quantity demanded of a particular good or service as a result of changes to other economic variables, such as its own price, the price of competing or complementary goods and services, income levels, taxes.
Demand elasticities can be influenced by several factors. Each of these factors, along with the specific characteristics of the product, will interact to determine its overall responsiveness of demand to changes in prices and incomes.
The individual characteristics of a good or service will have an impact, but there are also a number of general factors that will typically affect the sensitivity of demand, such as the availability of substitutes, whereby the elasticity is typically higher the greater the number of available substitutes, as consumers can easily switch between different products.
The degree of necessity. Luxury products and habit forming ones, typically have a higher elasticity.
Proportion of the budget consumed by the item. Products that consume a large portion of the consumer’s budget tend to have greater elasticity.
Elasticities tend to be greater over the long run because consumers have more time to adjust their behaviour.
Finally, if the product or service is an input into a final product then the price elasticity will depend on the price elasticity of the final product, its cost share in the production costs, and the availability of substitutes for that good or service.
Prices are also forecast using an input-output framework. Input costs have two components; labour costs are driven by wages, while intermediate costs are computed as an input-output weighted aggregate of input sectors’ prices. Employment is a function of output and real sectoral wages, that are forecast as a function of whole economy growth in wages. Investment is forecast as a function of output and aggregate level business investment.