Artificial Intelligence & Biopharma R&D IT
This article discusses issues around artificial intelligence & Biopharma R&D IT, and outlines how AI can be a potentially transformative technology for the biopharmaceutical and healthcare industries.
"...R&D productivity remains an ongoing concern. Artificial Intelligence and the accompanying analytics are now so advanced that these tools promise to improve the traditional drug target selection and R&D process." As noted in the Ernst & Young report Beyond borders Biotechnology report 2017 (1).
The PRISME Forum is the biopharmaceutical industry R&D IT leadership group that meets twice a year. It addresses common industry challenges, shares use cases and catalyses more rapid creation, adoption and application of solutions to increase the efficiency and effectiveness of biopharmaceutical R&D.
As such, there is a contribution that the PRISME Forum should make to the development and implementation of AI to reduce the time and cost of bringing new medicines to market to treat unmet patient needs. With that in mind, the PRISME Forum focused its Fall 2017 Technical Meeting on the potential of Artificial Intelligence (AI) to improve biopharma R&D and healthcare.
Definition of Artificial Intelligence
What is artificial intellence?
A google search reveals a practical definition, ie “the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making and translation between languages” (2). However, IBM Research provides a more pragmatic definition of AI: “By AI we mean anything that makes machines act more intelligently” (3) and this is the definition that will be adopted in this article.
There are many technical definitions and taxonomies that can be used to stratify the various computer science tools that live under the umbrella term AI. Indeed, AI experts can get excited about the nuances between terms such as machine intelligence and human intelligence or deep learning methods and pattern matching.
For further background, Stanford University’s initiative on the One Hundred Year Study on Artificial Intelligence or AI100 (4) provides a perspective on the history and future of AI.
This article identifies the start of the ‘AI100 timeline’ with Alan Turing’s 1950 paper on Computing Machinery and Intelligence (5).
Technologists working in Life Science R&D should be interested in the practical applications within Biopharma R&D. IBM’s pragmatic definition of AI is helpful and helps identify the opportunities in automation that are not necessarily in alignment with precise, computer-science definitions of AI.
AI & Machine Learning at a tipping point
AI and Machine Learning have reached a tipping point. While many of the AI-based, computer science techniques have been available for some time (eg neural networks) the increase in data availability, vastly more powerful and easily accessed computer power (eg GPUs in the cloud), increased interest in data science and analytics, and improvements in algorithms have transformed the AI landscape.
A recent Harvard Business Review Article (6) states that “In the sphere of business, AI is poised to have a transformational impact, on the scale of earlier general-purpose technologies… Although AI is already in use in thousands of companies around the world, most big opportunities have not yet been tapped.”
In addition, a recent Forbes article (7) cites “The necessity to start embracing AI technologies and revamping human resource strategies to create data science-driven interdisciplinary teams has become a matter of the future business sustainability for biopharma organisations.”
Biopharma opportunities in AI
A crowd-sourced, multi-author paper entitled ‘Opportunities and Obstacles for Deep Learning in Biology and Medicine’ (8) was created by a broad list of contributors (many academic) from around the world. The paper highlights a number of opportunities and challenges in applying deep learning to biology and medicine. It examines applications of deep learning to a variety of biomedical problems in particular: (i) patient classification, (ii) fundamental biological processes and (iii) treatment of patients. The conclusion was that the future would see “deep learning powering changes at the bench and bedside with the potential to transform several areas of biology and medicine”.
Discussion at the meeting revealed that many technology-based companies were highlighting their AI capabilities as the market quickly responded to the enthusiasm for the potential of AI, and in particular to the use of AI in life science R&D and healthcare. More than 30 relevant examples (see Table 1) were quickly identified, but it was widely recognised that any list of organisations active in the evolving R&D IT/healthcare AI landscape would be rapidly evolving.
Table 1 illustrates that there are many opportunities for the application of AI across the life science R&D/healthcare pipeline. However, the timescales in which they might create benefit varies. Importantly, there are today many nearterm opportunities for AI that are not as yet adopted broadly, but that have clear benefits for biopharmaceutical R&D. Examples might include:
- Image analysis and phenotypic screening of drug candidates.
- Drug repositioning and competitive intelligence through data integration.
- Clinical trial cost and timeline improvement (eg protocol authoring, patient recruiting, site monitoring and risk assessment have already been implemented in commercial products and services from CROs).
- Cost savings in pharmacovigilance and regulatory reporting (eg through Robotic Process Automation [RPA]).
What are the implications of Artifical Intelligence for the biopharmaceutical industry?
The 60 or so participants at the meeting were organised into five different discussion groups to consider the actions that biopharmaceutical companies would need to take to be in a position to derive advantage of this emerging AI technology. Each group considered one of the following five perspectives viz: Skills, Data, Organisation, Infrastructure and Metrics.
Leveraging AI in biopharma required the IT function, and the R&D IT groups, to have a strong foundation of the traditional skills and the willingness, flexibility and capability to acquire new skills. As illustrated in Figure 1 these skills span science, mathematics and technology.
There are also foundational capabilities required that are cultural and cross-disciplinary and include a process to innovate rapidly and to assess value from ‘placed bets’ in the quickly-changing AI landscape. The MIT Sloan article entitled ‘How Innovative is Your Company’s Culture’ (9) provides practical guidance for assessing a corporate culture and highlights that an innovative culture rests on a foundation of six building blocks, viz: resources, processes, values, behaviour, climate and success.
The adoption of AI follows a similar trajectory as with other technology innovations (reference the Gartner Hype Cycle (10)). There are many examples and articles on innovation management processes which generally focus on:
- Knowing what problem it is one needs to solve.
- Establishing success criteria.
- Experimenting rapidly – either to fail quickly or to demonstrate value.
- Scaling up successful experiments.
The Global Innovation Management Institute describes one ‘Rapid Iterative Experimentation Process’ (RIEP – pronounced ‘reap’) in the article ‘Rapid, Iterative Experimentation Process – a Lean Startup-style Approach to Innovation’ (11). Sanjoy Ray at Merck has been an influential voice on this topic in pharma, describing a hypothesis-driven approach and “rapid short experiments where ‘good failures’ are celebrated” (12).
The recent boom in Big Data, and the development of the management principles to exploit Big Data, is one of the driving forces for the re-emergence of AI. Without Big Data (preferably of high quality and lots of it), AI would be starved of the raw material upon which it depends. Four key areas of focus for biopharma successfully to adopt AI capabilities were highlighted, viz: (i) Data Strategy, (ii) Data Governance, (iii) Knowledge Representation and (iv) Data Stewardship.
(i) Data Strategy focuses on establishing the overarching strategy for how a company wants to create, manage and use its data assets. A Data Strategy provides a roadmap for a company to advance its data capabilities, and includes addressing key questions such as:
- What is the data that will generate competitive advantage, that we should maintain for the enterprise over time?
- What data do we need that is external to our company?
- What data do we need to generate to be competitive?
(ii) Data Governance is comprised of the overall management of the FAIR data principles (findability, availability, integrity, reusability) along with the security and confidentiality of the data used in an enterprise. A good Data Governance approach defines:
- Who is the responsible owner for data.
- What quality standards should be applied to data.
- How data will be managed to those standards going forward.
Data privacy regulations (eg the European Union GDPR) are an important topic. Organisations need to ensure that they were knowledgeable about the regulations and were in a position to be in compliance when the GDPR came into force in May 2018.
(iii) Knowledge representation focuses on:
- How we store and represent data.
- How we structure data for usability.
(iv) Data Stewardship focuses on the skilled resources required to put a Data Strategy and Data Governance into practice. It addresses the challenge of how to resource and operate on a daily basis the curation and maintenance of an organisation’s data and the continuous improvement of data quality and value.
Intellectual property and data pose new challenges in the context of AI. As an industry, the following questions arise: (i) When are we able to share pre-competitively? (ii) When are we willing to share pre-competitively? (iii) When does sharing put intellectual property at risk?
The biopharmaceutical industry has, in general, considered knowledge of software and technology to be pre-competitive and sharable, while data and information about specific compounds, and the specific processes by which they are developed into drugs, is considered proprietary and not sharable.
The question is: with an algorithm trained on proprietary data sets, would biopharma companies be willing to share the algorithm independently of the data? For many companies, the answer may well be ‘no’. If the data cannot be shared, then the algorithm trained on that data cannot be shared. However, the untrained algorithm could potentially be considered sharable.
AI thrives on training data sets. The richer and more diverse the training data set, the richer and more diverse can be the interpolation from the AI machine. Cross-industry opportunities to share data can create richer training data sets and allow AI algorithms to function better. The IMI e-TOX Project (13) provides one example of such a pre-competitive, collaborative, data-sharing initiative.
With the adoption of cloud-based technologies, the role of the IT organisation in the enterprise is evolving rapidly. The boundary between IT and other functions is changing as IT becomes a facilitator and more deeply embedded within business functions. This changing landscape comes even more sharply into focus as new skills and so-called ‘double deep’ resources (ie personnel knowledgeable in both business functions and the technology) are required to leverage data and the underlying technologies of AI.
There is no simple answer to the question of whether an enterprise should adopt a centralised or a decentralised approach, eg a single Chief Data Officer and Centers of Excellence versus a decentralised approach with data scientists embedded in the functions. Success depends instead on many factors ranging from company size and culture, to process, technology and data maturity.
Companies vary in the way they support technology innovation such as AI. Regardless of model, centralised or decentralised, successful AI technology implementation requires a focus on cross-functional partnership and teams. While the lines of responsibilities may be blurred, there are primary responsibilities that are clear and a hybrid model with platforms and centralised skillsets serving embedded domain experts is illustrated in Figure 2.
Platforms and data standards should be maintained at an enterprise level by IT organisations and leverage IT skillsets such as information architecture, systems architecture and knowledge representation. Data reporting and analysis should be the focus of embedded experts in the sub-disciplines applying AI. It can be argued that both centralised and decentralised approaches might increase the pace of adoption.
For example, centralisation encourages the recruitment and retention of scarce skillsets, creates standards, shares use cases across functions, increases organisational commitment and the faster reuse of technology. However, decentralisation increases discipline agility and expertise, and discipline-level innovation keeps focus on value and with a lower governance hurdle, enabling faster deployment of resources.
Many capabilities founded on AI evolve into something else as they become successful. Examples might include help desk automation, image analysis, imaging biomarkers, supply chain analytics, genomic data analysis, computational chemistry, field force and promotion analytics, clinical protocol authoring. By the time the system is working well, responsibility for using the technology is more amenable to decentralisation.
An IT organisation should strive to create the environment for domain experts to self-serve their reports and analytics and to be a partner for advice and implementation of new platforms and technologies or to manage new types of data. This ensures less siloing of data and redundancy of platforms and builds enterprise level data assets.
AI infrastructure requirements fall into two categories: (i) The foundational capabilities that most companies have today which might need enhanced capability or greater agility or more mature processes to be effective for AI management (shown in grey in Figure 3), and (ii) the emerging capabilities that are not yet as mature in most companies today and will require new investment to create (shown in blue in Figure 3).
Much of the infrastructure cannot be achieved if the enterprise does not first have a clear data strategy and data inventory as part of the data management environment.
What metrics will be most effective in measuring AI maturity and the success of AI implementation efforts? There are at least three broad categories for measuring the success of AI efforts:
Experiments: A number of AI pilots, and within these pilots there should be cycle time reduction, cost savings or quality metrics for rapid assessment of value. There should be the rapid conclusion of pilots (both successes and failures) and then for production implementation of the successful pilots.
Automation: Where we should see a high percentage of the process becoming automated and human responsibilities becoming concentrated on the higher-level, decision making, organisation of personnel and scientific-interpretation tasks.
Decision quality: Metrics that show that AI enabled decision-making increases decision quality.
Setting accountabilities and goals around these areas can provide the organisational incentives successfully to advance the AI agenda.
Summary - Artificial Intelligence and Biopharma R&D IT
AI is a potentially transformative technology for the biopharmaceutical and healthcare industries and has many applications for which a rapidly evolving set of technology vendors and services is emerging.
The PRISME Forum Technical Meeting in November 2017 brought together R&D IT experts from across 30 top biopharma companies to map a path to successful adoption of AI, and to prioritise the actions that organisations should take to be competitive with artificial intelligence and biopharma R&D IT . There are implications for all facets of R&D IT, viz: Skills, Data, Organisation, Infrastructure and Metrics.
Of these, staffing with the right set of skills, redefining the role of IT and measuring success are the most important factors in equipping a company to obtain competitive advantage with AI. New capabilities must be built on a solid R&D IT foundation of people, process, culture and technology. Strong partnership and cross-functional teams that bridge organisational constructs are essential to successful innovation. DDW
The PRISME Forum (http://www.prismeforum.org) is the de facto R&D IT leadership group of the global biopharmaceutical industry. Currently it has more than 40 members drawn from 30 different biopharmaceutical companies representing nine of the top 10 companies, and 25 of the top 30 companies by R&D expenditure. The PRISME Forum’s mission is “to enhance the efficiency, effectiveness and impact of global” R&D IT in biopharma. Each PRISME Forum hosts a Technical Meeting which brings together relevant opinion leaders and technical experts to increase the awareness of the PRISME Forum members of the opportunities for biopharmaceutical R&D IT to advance patient-centred, drug discovery and development by optimising R&D IT. For further details contact John Wise, the PRISME Forum Programme Co-ordinator, at jcmwise(at)prismenforum.org
1 Giovannetti, Glen T et al (2017). E&Y – https://www.ey.com/Publication/vwLUAssets/ey-biotechnology-report-2017-beyond-borders-staying-the-course/$FILE/ey-biotechnology-report-2017-beyond-borders-staying-the-course.pdf.