Big data and AI offer massive opportunities to the pharmaceutical industry — in theory. In reality, many companies are struggling to realize the potential of these tools. Some organizations have been hesitant or resistant to leveraging the technologies. Others may have attempted to embrace them early on but are now beginning their second or third incarnations of “digital transformation,” likely with some layoffs along the way.
Why the difficulty? Digital transformation is, of course, a massive undertaking — requiring enterprise-wide coordination and a clear, focused vision. In the real world, organizations have struggled with defining a focus for their AI efforts and sustaining the investments necessary to reach them. It’s easy to get excited about the prospect of using AI to solve everything under the sun, but more often, successes are coming when teams stay focused on very practical, targeted applications.
Despite the turmoil, the success stories show that when organizations set out with the right plan and tools, AI can fuel extremely valuable breakthroughs. Here are three factors to success.
Let go of legacy systems and the concept of “my” data
Legacy infrastructure and data silos are major hindrances to AI research. While it is true that replacing a legacy storage system with modern architecture is a major undertaking, older systems simply aren’t built to support the data standardization and collaboration that is needed for AI development. AI is a data-hungry tool, and pharma organizations must leverage all the data at their disposal to create accurate models. This means embracing systems suited to deal with big data and collaboration, with tools to support compliance, standardization and discoverability. Adopting standards and platforms early on to unite data across modalities can dramatically accelerate the process.
In some organizations, especially global ones, there is a mindset of “my data” and “your data” versus an enterprise-wide culture of “our data.” This often relates to the infrastructure issue and the challenges of accessing and reusing data with legacy systems. If data is stored locally or archived in other inaccessible ways, it has no chance of contributing to the organization’s larger success.
Keep a narrow focus, and consider ROI early on
There is a sentiment among many in the pharma industry that AI hasn’t entirely delivered on its promise yet. However, it may be more accurate to say that organizations haven’t quite figured out the appropriate problems for AI to solve. Instead of approaching AI with an expectation that it can solve everything, we must adopt more systematic and organized ways of leveraging the technology to address specific types of problems with outcomes that can be measured.
Under this approach, an organization might define a specific scientific question or business need and then work backward to determine if AI is an appropriate tool to help answer it. These questions might be in areas like patient outcomes, image analysis, or cohort design — all examples that are narrow enough to pinpoint the data types necessary to build a successful AI model. Once the problem is defined, it must be determined if the organization has that data in the appropriate volume and diversity and whether it can be properly curated for AI development.
Teams should also have a clear goal at the outset of their AI project in terms of how the results will be applied and the potential ROI of those applications. If they are intended for internal use, what is the potential impact of the technology on increased efficiency, waste reduction, forecasting, etc.? And if the AI is intended for commercialization, other preparatory steps are necessary to ensure compliance and increase the chances of a smooth path through regulatory approval. These steps must be considered at the beginning of the project rather than treated as an afterthought.
Recruit senior-level champions
Often, leaders of AI initiatives must educate upward to ensure that organizational leadership is on board and that they understand the timelines and investment required. As discussed, AI efforts demand enterprise-wide collaboration, so senior-level support is critical to enforcing standards and keeping the initiative an ongoing priority. Enlisting executive support will be much easier if the work has been done to define a clear focus and potential ROI for the effort.
With help from the top, a clearly defined business case, and modern data management tools, researchers are in a much better position for successful AI projects that can drive exploratory analysis and accelerate drug development.
Costas Tsougarakis is vice president of life sciences solutions for Flywheel, a biomedical research data platform. Costas has more than 20 years of experience in research, development, and delivery of scientific information systems in the medical device, clinical research and pharmaceutical domains with an emphasis on biomedical imaging. Prior to joining Flywheel, he led the development of the Roche Global Imaging Platform, based on FAIR principles, which enables at-scale image data storage, ingestion, curation and analysis.
Filed Under: Data science, Industry 4.0, machine learning and AI