Guidelines from Agnostic Methodology Applicable to All AI Algorithms
SAN FRANCISCO, CA – (August 18, 2020) – Aizon, Inc. (Aizon), the leading manufacturing artificial intelligence (AI) and advanced analytics platform company for regulated industries, today announced that a new study, published online in the Parenteral Drug Association’s Journal of Pharmaceutical Science and Technology (PDA JPST), demonstrates that AI algorithms can be qualified for pharmaceutical product and medical device productivity chains. Qualification was achieved using Aizon’s GxP-compliant AI software-as-a-service (SaaS) platform.
AI algorithms are extremely valuable to the highly regulated life sciences and healthcare industries because they can be used to make critical decisions during drug and medical device manufacturing processes. However, AI algorithms must first be qualified to ensure that they enable established manufacturing goals. Until now, such a procedure did not exist in GxP environments.
“Our study specifically qualified the Isolation Forest outlier detection algorithm, but it did so from an agnostic perspective, which enables the resulting guidelines to be abstracted to other AI algorithms for regulated drug and medical device manufacturing environments,” said Toni Manzano, Ph.D., Chief Science Officer and Co-founder of Aizon, and principle study investigator. “Now, for the first time ever, AI algorithms can be qualified and applied by AI models as the foundation for decision making to leverage productivity and quality for critical processes in pharmaceutical and medical device manufacturing.”
Manufacturing decision making that is informed by qualified AI algorithms can potentially lower costs and increase profitability related to pharmaceutical and medical device production. Leveraging AI as a multivariate tool can help optimize processes, as well as predict and fix anomalies in industrial environments. As a result, products and devices that are safer, more efficient and effective, and of higher quality can be made accessible to patients and healthcare providers.
The study used a Quality by Design (QbD) approach and data generated from equipment called bigBox, which was designed specifically for the research by Aizon. Real-time data related to four parameters (MS Main, MS Brake, MS Misalignment, MS Imbalance) was relayed from bigBox to Aizon’s GxP-compliant, AI SaaS platform every five seconds. A structured Design of Experiment (DoE) was performed using all the operational ranges of bigBox, as well as outliers in the experiments, since the algorithm being qualified was Isolation Forest. Use of QbD allowed characterization of the algorithm and identification of which designed space areas built into the synthetic data sets performed better. The devised DoE and experiment strategy resulted in a valid qualification for the Isolation Forest algorithm.
The study, AI Algorithm Qualification, which is currently available online, will be published in the January/February 2021 issue of PDA JPST.
About Aizon, Inc.
Aizon is a software provider that transforms manufacturing operations with the use of advanced analytics, artificial intelligence, and pharma 4.0 technologies focused on optimizing pharmaceutical and biotech companies. The Aizon analytics platform seamlessly integrates unlimited sources of structured and unstructured data to deliver actionable insights across all manufacturing sites. Aizon offers an intuitive way to gain meaningful operational intelligence with data by enabling real-time visibility and predictive insights in a GxP compliant manner with end-to-end data integrity. Founded in 2014, the company is based in San Francisco, California and also has a European office in Barcelona, Spain.
About PDA JPST
PDA JPST is the primary source of peer-reviewed scientific and technical papers on topics related to pharmaceutical/biopharmaceutical manufacturing, sterile product production, aseptic processing, pharmaceutical microbiology, quality, packaging science, and other topics relevant to PDA members. PDA JPST is an internationally recognized source that receives over a quarter of a million visitors annually.
