ParticleSentryᴬᴵ is a software product that brings artificial intelligence and particle imaging analysis technologies together in biopharmaceutical formulation development, scale-up and production. Our technology combines machine learning and computational statistics to identify protein aggregation and support biopharmaceutical characterization, enabling better understanding and control of your drug product. Our patent-pending supervised learning technology creates "fingerprints" that quantitatively characterize protein drug production and protein aggregates under different stressors. These fingerprints are used to identify and resolve issues at every stage of product development and manufacture of large molecule protein therapeutics.
ParticleSentryᴬᴵ is compatible with flow-imaging microscopy, backgrounded membrane imaging and microfluidic imaging platforms, and employs a convolutional neural network to perform particle image analysis and outlier subvisible particle detection. When used for biopharmaceutical characterization during formulation and development, the fingerprint of natural state drugs lays the foundation for heterogeneous particle analysis. Subvisible particle images taken from drug stressed states like those encountered during tangential flow filtration, mixing, freeze-thaw, pH and viral clearance, pump/filling, shaking and light induced stressors expand the utility of the fingerprint by enabling comparison of future outlier results and speed root-cause analysis. The fingerprint can also be valuable in analysis of container/closure, timed storage, forced degradation activities, and follows the drug into scale-up and production.
Combining Machine Learning and BMI: A Case Study 06/2022 (pdf)Download
Testing Precision Limits of Neural Net-Based Quality Control Metrics in Digital Microscopy 02/2022 (pdf)Download
Machine Learning and Accelerated Stress Approaches to Differentiate Causes of Aggregation 03/2021 (pdf)Download
Machine Learning & Statistical Analyses for Extr. & Char. Fingerprints of Antibody Aggregation 04/20 (pdf)Download
Shifting Paradigms Revisited: Biotechnology and the Pharmaceutical Sciences 08/2019 (pdf)Download
Deep CNN Analysis of FIM Data to Classify Subvisible Particles in Protein Formulations 12/2017 (pdf)Download
Light obscuration, the USP <787>, <788> compendial method for subvisible particle analysis, encounter difficulties when applied to proteinaceous drugs. This presentation covers how we combine particle image analysis, deep machine learning, computational statistics, and high-throughput microfluidic imaging to characterize subvisible particles and protein aggregates in biologic drug products. By analyzing the morphological and textural particle features contained in the image, this method delivers quantitative, actionable information for formulation development, container qualification, and fill-finish quality control. Four case studies are reviewed. .
This video demonstrates the "Create Fingerprint" functionality provided in ParticleSentryᴬᴵ software.
This video demonstrates the process steps to compare new sample images to an existing fingerprint model in ParticleSentryᴬᴵ software. The video also briefly demonstrates the process to view the subvisible particle images contained in your test.
Normal Process Fingerprint vs. Process Shift Fingerprint
You need a large collection of subvisible particle images, the more images the better. If you have known reference condition images and images from related accelerated stability studies, they can be incorporated into the fingerprint as well. More images enhance the overall particle image analysis. Images can be collected from historical tests or newly captured.
The output includes a fingerprint model (probability density function), confusion matrix, mean and standard deviation values, and P-value for the test statistic, and other visualization tools to diagnose when process upset occurs.
Definitely. As new data is captured and understood, new image conditions can be added, and a new fingerprint model (probability density function) can be generated containing the new class(es). This allows for easy root-cause analysis and issue/outlier detection in the future.
ParticleSentryᴬᴵ software is large molecule specific- so any biologic formuation is within our particle image analysis capability. In our experience, formulations differ and are significantly influenced by factors like excipients, surfactants, etc. While much of our work has focused on mAbs, we are not constrained to look only at mAbs.
Our technology will work with any modality that generates an image (e.g., background membrane imaging (BMI), microfluidic imaging (MFI), flow imaging microscopy (FIM), fluorescence, Raman, imaging flow cytometry, etc.).
Yes. While the bulk of our work has been with flow imaging microscopy, we do have experience with other modalities.
Particle size and count data from light obscuration (LO) has been proven inaccurate when analyzing protein-based drugs. LO cannot accurately resolve the translucent particles found in protein drugs and undercounts or under-sizes these particles. Protein drugs are inherently aggregation-prone and manufacturing stressors increase the likelihood of aggregate formation throughout the manufacturing process. Many drug manufacturers already use flow imaging microscopy, backgrounded membrane imaging and microfluidic imaging as orthogonal measurements to LO particle size and count data. Historically, these orthogonal methods have not leveraged the actual images. ParticleSentryᴬᴵ software leverages the rich textural and morphological detail of the subvisible particles and protein aggregates. These data allow you to move beyond particle size and count alone and characterize your drug product - as shown in the fingerprint model (probability density function) images. This quantification allows easy detection of new outlier conditions. A robust fingerprint model includes many process issues (pH shock, pump failure, etc.) and provides valuable data for quick root cause analysis. Finally, part of the regulatory burden placed on biologic manufacturers involves demonstrating an aggregate control strategy throughout the development and manufacturing processes, and ParticleSentryᴬᴵ software, in addition to providing actionable data, delivers on the requirement to demonstrate an aggregate control strategy.
YES! One of the key use cases for ParticleSentryᴬᴵ software is monitoring at-line or on-line at various points in the manufacturing process. Excipients, surfactants, and other additives can change the morphology and composition of subvisible particles and protein aggregates, and are typically studied during candidate selection and formulation. ParticleSentryᴬᴵ software, when deployed during formulation development and preclinical trials provides fingerprint models of the particle images created when ingredients are used in formulation testing and development. These fingerprint models can be compared to subvisible particle and protein aggregate images expected during manufacturing of biologic drug substances at fill finish.