Data Transformation, AI, and Data Platforms That Work

May 13, 2022

May 15, 2022 | Chris Dagdigian wasn’t the only one reporting from the IT trenches last week. The Bio-IT World Conference & Expo’s annual Trends from the Trenches panel also featured three other BioTeam consultants: Adam Kraut, Michelle Bayly, and Anna Sowa. The three reported on their own observations in their capacities as BioTeam consultants: focusing on leadership trends, platform best practices, and AI.

Data Trends for Leaders

Adam Kraut reported on leadership trends on bio-IT C-suites. “Data transformation”, he said, is the current leadership umbrella term for big data, cloud, internet of things, and AI. Data transformation is really applying a DevOps mindset to your digitized organization, he said. “What did the DevOps movement teach us? It told us that it’s not about the technology and processes, it’s actually about the people, the culture, and changing the way we work with each other,” Kraut pointed out.

With this in mind, Kraut dismissed any lament over data transformations “failing”. Continuous improvement is the goal, he said, not a single leap forward. Instead, companies should focus on building healthier habits to improve digital health and yield a strong data ecosystem. Kraut uses “data ecosystem” to mean, “the set of infrastructure and services that empowers a community of data scientists and engineers to make decisions and influence business outcomes.” The data ecosystem also includes the people, he said, and the culture of data generation and use at a company.

Kraut explored hallmarks of a healthy data ecosystem that included data discoverability, integrity at the origin, a culture of data citizenship, use of common languages and standards, leveraging automation, tracking experiments, and a continuous delivery mindset.

Humans love to create silos, he pointed out; it’s natural for us. But data discoverability is essential to maximizing the data science, machine learning, and AI that a company can do. He called for well-defined metadata and curation at the point of data instantiation and registration with experiment design, analysis-ready data, and abstract data from storage services.

Agility should be a differentiating capability for a company, not being married to a particular set of tools. Infrastructure can be viewed as code, he acknowledged, but “beware the tradeoffs of deciding to go down the path of automation. If the design of the automation or doing that automation doesn’t give you the return on investment, then you probably shouldn’t do it. Make it manual, and only put in the effort to automate if you’re actually going to get that return on investment,” he said. He also challenged companies to solve their personal problems, not simply copy other companies or players because it sounds good.

Finally, he challenged companies to apply the DevOps mindset—quickly deploying solutions in the real world—to machine learning and AI. Currently there is a gap between nascent AI algorithms and real-world applications, Kraut said. “That massive gap is going to require extremely competent cross-functional teams. We’re going to have to eliminate mobile hand-offs and all the kinds of stuff that DevOps taught us. Automation will be fundamental,” he said.

Kraut also flagged a leadership shift from a process to a product mindset. He quoted Mason Victors from Recursion Pharmaceuticals, who told Bio-IT World that the pharma invests in systems to generate, characterize, and analyze data as fuel, not exhaust. Viewing data as a product engenders a sense of discipline and ownership, he said, and emphasizes the dynamics between data producer and data consumer. Overall, Kraut advocated for a bias toward action. Move fast, he said, and try more things. When you try more things, you’re more likely to find ones that work for you.

Finally, he highlighted the characteristics of leadership that is ready to empower agile digital transformation. These leaders are diverse and focused, both leaders and humble followers when required. They own their work, but also empower others, nurturing and mentoring when appropriate.

Build With the User in Mind

Data warehouses, data lakes, data soup—Michelle Bayly is over the “magical” descriptors. “I guarantee you, that if you tell me that you’ve got a data BLANK, and I ask five other people in your organization, they will call it something else,” she said.

Going with “data platforms”, Bayly outlined the steps needed to get to the ideal: a low-cost, easy to maintain, FAIR source of fresh, quality data that can be used by AI and machine learning systems while also easily accessed by researchers, clinicians, and leadership.   

First, she challenged companies and organizations to be honest about the status quo. She asked: What is your use case? Who are your users? Do you have the infrastructure you need? Is your data management a hot mess? (This one she answered: Yes!) What is your timeline and what phases are you expecting? What is your change management—and conflict resolution!—plan? What can you afford?

Bayly warned that rushing into data platform builds often leaves some of these questions ignored, resulting in platforms that work for some (developers) but not others (researchers). Perhaps you don’t have a plan to scale, or perhaps you have over-built a solution that is far more complex than the use case demands. Foundational concerns like getting to know your data and intimately understanding your use cases should all happen before you choose or build an architecture.

Like Kraut, Bayly emphasized the need for a people focus. Software will never replace the need for human data management, she argued. Budget for it. And don’t build the system to suit IT.

“You need to stay focused on that user,” she said. “Keep coming back to those use cases before, during, and after you build. I promise you, if you do not, they’re still going to go back to using Excel and to transferring USB drives in parking lots.”

Feasting on Low Hanging Fruit

Anna Sowa dug deep into AI as the final BioTeam panel speaker. AI is both underrated and overrated, she argued. No, AI isn’t going to cure cancer, she said. “But why are we asking AI to cure cancer? There are so many awesome things that ML can do for us in our labs right now that we’re not taking advantage of.”

AI is inherently social, Sowa argued, and will mirror both our biases and our ethics. As such, she warned against “AI cowboys” who claim to be able to ride in and save you, arguing instead to start with existing teams that care about—and understand—the data and processes already in place. Creating a data commons is a team building exercise, she said. Bring together a diverse team with different backgrounds, talents, perspectives, and expertise. Give them the time and support they need to stretch into new positions and bring their institutional and data knowledge with them.

Then think about AI rightly, she said. AI is useful, but not a crystal ball. Ask narrow and meaningful questions. Celebrate what’s achievable now and allow AI to explore the boundaries between what we know and what we don’t know.

She acknowledged that moving into AI might be a slightly painful process; make it beneficial for teams. She advised that CEOs take responsibility for driving the transformation. Putting if off will be far more painful than learning as you go. Every transformation needs to start somewhere, she challenged. Break the first silo.