December 14, 2022
by Andrew Anderson, Vice President of Innovation & Information, ACD/Labs
A new year is historically a time for reflection. In 2022, ACD/Labs has been in the scientific software industry for 28 years. As I think about our commitment to solving customer challenges with software, I’m reminded how important it is to deliver solutions that make our end users—scientists—more efficient, so they’re supported in the decision-making process. We believe that to create value for a customer, you need to intimately understand the customer’s workflows, processes, data flows, and overall business and IT goals.
In a recent discussion with our CEO Daria Thorp, she reflected on ACD/Labs’ progression: “In the early years of ACD/Labs, we became known for chemistry software that gives scientists predictive insights and removes the tedium of data processing, analysis, and reporting. We were supporting innovative decisions and boosting productivity of individual researchers, many continue to be our users to this day. ACD/Labs’ present is in combining those productivity tools with automated knowledge management and decision support (including, for high throughput experimentation), and its efficient dissemination to multifunctional and geographically distributed teams. Today, whole teams and CMC groups use our software to facilitate their data-driven decisions. ACD/Labs future is being shaped by the evolving needs of the chemical and pharmaceutical R&D industry that strives to get the most out of their data.”
Three key topics have come up repeatedly in conversations with customers in 2022 that I find notable for our work ahead.
1) Scientist-centric data use vs. machine-access
Over the past few years, organizations’ mindsets have shifted from associating scientists as the sole endpoint for using data, to understanding that scientists and machines need the data. Today, organizations want to facilitate end-user scientist and machine access to improve efficiencies and processes beyond the project level. The requirements for these two use cases are usually not fully met with current workflows and data management practices. Bench scientists and data scientists have different requirements for data visualization, readability, normalization, etc. Well-structured data is a huge benefit to organizations because it becomes more accessible to humans, and may be used to train predictive applications.
It usually requires different analytical techniques to gain a complete characterization of a substance’s composition. Every substance has an identity, usually confirmed through mass spectrometry, NMR, and other spectroscopic means. Taking it a step further, chromatography can be used as a complementary technique to characterize the substance’s composition, creating even more value. When scientists can provide machines with training data about composition and substance, they can characterize that substance further. So, if a scientist can have a digital and physical representation of the synthetic, formulated, or biological processes, they have cause and effect. The cause would be specific conditions that produce a particular output. That output, typically a physical material, can then be categorized chromatographically and spectroscopically to produce compositional info. As scientists collect a lot of information about input and output, a digital representation of a synthetic process makes it possible to leverage machine learning (ML) to predict reaction conditions, produce an identity, and more.
Scientists need to be able to access that data and decipher it, but it also needs to be machine-readable. If scientists have collected experimental information in an ELN, they can use applications on the Spectrus platform to capture the digital representation of composition and identity.
This leads very neatly into the next topic…
2) Machine learning is the answer, but what’s the question?
When organizations look to invest in new strategic capabilities, they often identify digital transformation as a key area of investment and prioritize ML in the process. This strategy and investment should be carefully considered because scientists and their organizations at large need ML to be able to do two core things:
- Glean insights from existing operations/data
- Use those insights in meaningful ways to promote change
Before investing in ML, organizations need to understand its explicit set of needs, data structures, and how to engineer pipelines to prepare the content. Laying out the strategy and goals beforehand can help determine what strategic capability investment to make. Whichever choice is made, organizations need to think realistically about how they will take a set of business processes and understand explicitly and exhaustively how and where those business processes produce data. Some questions worth asking should center on the data’s format and if it needs to translate into another format suitable for aggregation and assembly. There are many different types of data formats, including Allotrope’s ADF format, ASM format, ontologies, and other competing standard formats like AniML and JSON. In order for a ML initiative to take hold, especially in the analytical space, data standardization needs to occur.
Establishing ML as a goal is reasonable, but the hard work is looking at how that capability applies to your organization’s business processes. Unless the plan is mapped out where each step is detailed, it’s a rushed adoption. Strategic capabilities are certainly valuable, but it’s imperative to take the time to carefully manage your organization’s lift and adapt process for the particular interest.
In this scenario, we recommend that organizations prototype, pilot, and outline achievable objectives by tracing the work to measure the transformation. Keeping the chain of events connected and integrated can help ensure that the results and the process are not rushed. At ACD/Labs, we recommend taking an iterative, pilot-based approach as the first step. Then from the pilot experience, take a natural break in the transformation journey and learn from the prototype experience. This can help identify any risks to address before moving forward.
3) Data flows: Doing more with less
Analytical chemists continue to look for ways to run more experiments in less time while producing accurate and timely results. Chemists will continue to be interested in automating all types of workflows and data flows by seeking ways to marshal data from an instrument to see results and find a conclusion with limited scientific intervention.
In the future, analytical workflows will be automated to the point that a scientist can rely on accurate and correct calculations and computations with the algorithms in place. Additionally, scientists will only need to review data “by exception,” and all routine data processing can be automated and validated at once, allowing for significant time savings for an analytical scientist.
Like any technology, it needs to fit into specific use cases and solutions. When you combine automating workflows and data flows in the lab with the ability to harmonize and standardize all types of data, you allow the convergence of multiple technologies. Not only does this enable valuable solutions to be created in support of drug discovery and development, but it also helps in other aspects of clinical trials, real-world data analysis, and the general healthcare industry.
Providing valuable tools for our customers is the most rewarding experience for us at ACD/Labs, and we look forward to continuing this into the new year.
About the Author
