Pistoia Alliance Research Finds Concerns on Security, Barriers, More
By Bio-IT World Staff
October 16, 2024 | As artificial intelligence (AI) and machine learning (ML) become increasingly prevalent in the industry, there are still several concerns about its implementation. Mainly, how safe is it to use AI? What measures can be taken to increase and reinforce security? The Pistoia Alliance decided to find out.
Pistoia Alliance distributed a survey to 200 R&D experts from life science start-ups to large pharma companies across Europe, North America, South America, and the Asia-Pacific region. The survey was conducted online from May to August 2024 and includes questions aligned with a 2023 survey also examining priority investment areas, as well as the barriers to implementing technology and the benefits organizations expect to gain from digitalizing and automating R&D labs, says Becky Upton, president of the Pistoia Alliance.
Interest in AI/ML has, unsurprisingly, increased since the previous year. The survey found that 62% of respondents reported an increase in investments toward AI/ML over the next two years, and some 68% are already using AI/ML in their labs, which is an increase from 54% last year.
Improving the efficiency and effectiveness of R&D was seen as a top benefit of digitalizing and automating the lab, up from 71% to 76% since 2023. Furthermore, on a global scale, people have become more collaborative. Cultural barriers and institutional resistance to sharing data have dropped to 35% in 2024, which indicates that people are becoming more open to working together. However, 25% of respondents cited a lack of sufficient tools and systems as the biggest barrier preventing cross-lab collaboration. This is because the life sciences industry struggles to generate and maintain well-structured, interoperable data across labs and institutions, which cannot keep up with the speed of AI and ML development, Upton says. Many of these systems simply are not equipped for modern AI workflows, she added.
But the increase of AI and ML incorporation and implementation indicates a greater need for collaboration and knowledge to overcome data sharing challenges. Yet a significant number of respondents still hesitate to implement AI and ML at their companies. The survey reported 50% of respondents cited low quality and poorly curated datasets as the biggest barrier to AI implementation. More than 40% cited privacy and security concerns around data as another reason—an uptick from 34% in 2023. Upton states that the EU AI Act may help address these privacy and security issues. However, more awareness and clarity are required around how it applies and whether it can sufficiently offer guidance for ensuring both security and accessibility. More review and analysis are required to understand and digest the guide fully.
The third biggest barrier to AI/ML adoption is due to not having Findable, Accessible, Interoperable, Reusable (FAIR) principles applied to data, said 38% of respondents. “The big issue here is interoperability,” says Upton. AI models perform best when they have access to large, diverse, and high-quality datasets, and interoperability ensures that data from different sources (e.g. systems, institutions, and formats) can seamlessly integrate and be used together. However, due to the differences or insufficient metadata of each source, this means they are not FAIR. “Therefore, [they] cannot be integrated with other datasets or effectively used to train AI/ML models,” continues Upton.
More respondents are also calling for more data governance frameworks (49%), templates for standardization and metadata (46%), and best practices guides (45%) to move their labs forward. Furthermore, more than half of respondents (51%) asked for more maintenance and management of data standards and ontologies, and 29% report wanting ontology training.
It seems like education is a major key: 42% of respondents expressed interest in training courses on AI and ML in the lab. “As the use of AI becomes more prevalent in pharmaceutical research, it is important for all scientists to have a basic understanding of AI principles and applications,” explains Upton. She mentions that the Pistoia Alliance Lab of the Future Community is looking at developing a training program to increase AI literacy among non-data scientists. This program is intended to cover fundamental concepts, ethical considerations, and practical applications of AI in pharmaceutical research.