Informatics specialist ACD/Labs has announced plans to collaborate with Science Data Experts (SDE) to come up with ways for life-science organizations to make use of machine learning (ML) and artificial intelligence (AI) to drive R&D acceleration. Outsourcing-Pharma connected with two ACD/Labs representatives to learn more about the partnership, and how AI and ML can streamline data management:
- Andrew Anderson, vice president of innovation and informatics strategy
- Richard Lee, director of core technology
OSP: How did ACD/Labs and Science Data Experts come to work together—have you collaborated on projects in the past?
ACD/Labs: ACD/Labs and SDE recognized that the pharma industry is struggling to develop and implement AI and ML technologies into their workstreams. ACD/Labs and SDE have been working with these technologies for years and believe that, given the need and the nature of our joint experiences, we can bring significant value to the pharma industry by working together.
ACD/Labs has maintained an active collaborative relationship with SDE for the last few years, focused on respective core application development. Our companies’ respective expertise is in complementary areas: ACD/Labs for scientific and analytical data processing, interpretation, and management; SDE for ML modeling and data science in scientific applications.
We believe our respective customer bases will benefit from the companies working together in a synergistic fashion.
OSP: Please share a little more detail about what each partner brings to the table. I understand your areas of expertise are complementary, but what specialties does each of you offer?
ACD/Labs: Working with industrial scientists, both companies have identified that our customers are looking for predictive toolsets, which require highly curated, training datasets. ACD/Labs has been helping customers to more effectively marshal, contextually organize, and structure such data and content.
SDE provides the necessary details to data and data science strategy to help customers better leverage their data by building predictive tools (ML and AI), in addition to experience in data needs beyond data standardization and normalization (i.e., data format and disposition).
OSP: Please tell us about how the use of ML and AI have evolved in life sciences (especially in drug-development research) in recent years.
ACD/Labs: Indeed, life science ML exploration has evolved: from what was the art of the possible, to now seeing stakeholders focused on the art of the practical. Specifically, researchers are focused on applying the various AI and ML technologies to specific real-world applications.
From target discovery to the prediction of target activity modulating chemotypes, to ultimately predicting clinical outcomes, researchers are developing a variety of predictive tools to help accelerate better understanding of human health, but also accelerate the delivery of novel therapies to patients. Certainly, reducing the data engineering costs has become a key focus area; thus enabling the art of the practical.
Stakeholders have begun to focus on practical productivity applications of AI and ML, for example: accelerating development of stability indicating methods based on predicted degradation products. Traditionally, developing stability-indicating methods requires physical degradation under specific stress conditions and manual variation of chromatographic conditions until the most optimal separation is discovered.
Beyond analytical applications, typically most formulations are based on previously established compositions. ML/AI can assist in optimizing formulations to provide the best possible bioavailability, where the use of ML/AI could provide unexpected correlations.
OSP: How can advanced technologies like AI and ML accelerate drug discovery and development—what aspects of the process are made possible or greatly optimized with such tech?
ACD/Labs: One of the goals in discovery in using ML and AI is to predict clinical outcomes based on chemical input. As such, by reducing the number of physical synthetic and assay/performance testing required for candidate nomination, discovery timelines can be dramatically reduced.
In development, ML and AI can be applied to predict processes for drug substance manufacturing, drug product manufacturing, and corresponding control methods. By reducing the number of physical process development and validation experiments, one can thereby potentially reduce drug development timelines to produce sufficient quantities of clinical trial materials.
OSP: Please elaborate on how the ACD/Labs and SDE collaboration will help clients with:
- Synthetic or biologic process development and optimization
- Formulation development and validation
- Process development and validation
- Chromatographic method development and validation
- Leveraging analytical data to glean latent insights about processes, operations, and quality
ACD/Labs: Across these different application areas, ACD/Labs systems can help organizations identify, marshal, and produce well-curated, digital training sets. In areas 1–4 above, analytical data serves as one of the fundamental attributes of the work product of the experiments conducted.
Oftentimes such analytical data is abstracted or reduced to numbers, words, or pictures. By having full fidelity analytical data as a part of AI/ML training sets, unanticipated correlations in an artificial neural network (ANN) may be identified.
One benefit of more data and variables is the latent correlations for unsupervised ANN. Applications of AI/ML will be most beneficial when connections can be made from the automated ability to review data and apply it in unintended workflows.
OSP: What advice do you have for organizations who have not yet dived deeply into AI/ML—what should they know and understand before getting started in earnest?
ACD/Labs: Have a goal that is practical—be methodical in your approach and don’t attempt to look for panaceas that simply rarely exist. If you have a project with extensive scope you are more likely to fall short of your goals!
Our suggestion is to take a step back, prepare first, i.e., start with a journey map, look at the entire landscape, the goals of the projects, what data is available, where and how it can be assembled, along with the physical workflows you execute today that produce this data, then itemize the various formats and dispositions of that data.
Finally, establish traceability between the data and the formats required by your ML data scientists. Select the right AI/ML model and complete the implementation as part of a proof of concept (POC). If all goes well in the planning and execution, you should have a robust model that will help to accelerate your processes today and into the future
OSP: Do you have anything to add?
ACD/Labs: Starting with practical goals in mind is the best first step. Stakeholders should expect their AI and ML journey to be an iterative one. Our collaboration intends to help bridge the fluency gap between the interdisciplinary nature of AI/ML. We’ve certainly reflected upon the struggles to effectively adopt AI/ML in various industries; we believe that through our respective experience and capabilities, we can help our customers implement solutions and realize their aspirational productivity goals.