Supervised and unsupervised machine learning approaches for monitoring subvisible particles within an aluminum-salt adjuvanted vaccine formulation.

Suspensions of protein antigens adsorbed to aluminum-salt adjuvants are used in many vaccines and require mixing during vial filling operations to prevent sedimentation. However, the mixing of vaccine formulations may generate undesirable particles that are difficult to detect against the background of suspended adjuvant particles. We simulated the mixing of a suspension containing a protein antigen adsorbed to an aluminum-salt adjuvant using a recirculating peristaltic pump and used flow imaging microscopy to record images of particles within the pumped suspensions. Supervised convolutional neural networks (CNNs) were used to analyze the images and create "fingerprints" of particle morphology distributions, allowing detection of new particles generated during pumping. These results were compared to those obtained from an unsupervised machine learning algorithm relying on variational autoencoders (VAEs) that were also used to detect new particles generated during pumping. Analyses of images conducted by applying both supervised CNNs and VAEs found that rates of generation of new particles were higher in aluminum-salt adjuvant suspensions containing protein antigen than placebo suspensions containing only adjuvant. Finally, front-face fluorescence measurements of the vaccine suspensions indicated changes in solvent exposure of tryptophan residues in the protein that occurred concomitantly with new particle generation during pumping.

Effect of mechanical stresses on viral capsid disruption during droplet formation and drying.

Identification of the mechanisms by which viruses lose activity during droplet formation and drying is of great importance to understanding the spread of infectious diseases by virus-containing respiratory droplets and to developing thermally stable spray dried live or inactivated viral vaccines. In this study, we exposed suspensions of baculovirus, an enveloped virus, to isolated mechanical stresses similar to those experienced during respiratory droplet formation and spray drying: fluid shear forces, osmotic pressure forces, and surface tension forces at interfaces. DNA released from mechanically stressed virions was measured by SYBR Gold staining to quantify viral capsid disruption. Theoretical estimates of the force exerted by fluid shear, osmotic pressures and interfacial tension forces during respiratory droplet formation and spray drying suggest that osmotic and interfacial stresses have greater potential to mechanically destabilize viral capsids than forces associated with shear stresses. Experimental results confirmed that rapid changes in osmotic pressure, such as those associated with drying of virus-containing droplets, caused significant viral capsid disruption, whereas the effect of fluid shear forces was negligible. Surface tension forces were sufficient to provoke DNA release from virions adsorbed at air-water interfaces, but the extent of this disruption was limited by the time required for virions to diffuse to interfaces. These results demonstrate the effect of isolated mechanical stresses on virus particles during droplet formation and drying.

Features in Backgrounds of Microscopy Images Introduce Biases in Machine Learning Analyses.

Subvisible particles may be encountered throughout the processing of therapeutic protein formulations. Flow imaging microscopy (FIM) and backgrounded membrane imaging (BMI) are techniques commonly used to record digital images of these particles, which may be analyzed to provide particle size distributions, concentrations, and identities. Although both techniques record digital images of particles within a sample, FIM analyzes particles suspended in flowing liquids, whereas BMI records images of dry particles after collection by filtration onto a membrane. This study compared the performance of convolutional neural networks (CNNs) in classifying images of subvisible particles recorded by both imaging techniques. Initially, CNNs trained on BMI images appeared to provide higher classification accuracies than those trained on FIM images. However, attribution analyses showed that classification predictions from CNNs trained on BMI images relied on features contributed by the membrane background, whereas predictions from CNNs trained on FIM features were based largely on features of the particles. Segmenting images to minimize the contributions from image backgrounds reduced the apparent accuracy of CNNs trained on BMI images but caused minimal reduction in the accuracy of CNNs trained on FIM images. Thus, the seemingly superior classification accuracy of CNNs trained on BMI images compared to FIM images was an artifact caused by subtle features in the backgrounds of BMI images. Our findings emphasize the importance of examining machine learning algorithms for image analysis with attribution methods to ensure the robustness of trained models and to mitigate potential influence of artifacts within training data sets.

Stabilization of an Infectious Enveloped Virus by Spray-Drying and Lyophilization.

Enveloped viruses are attractive candidates for use as gene- and immunotherapeutic agents due to their efficacy at infecting host cells and delivering genetic information. They have also been used in vaccines as potent antigens to generate strong immune responses, often requiring fewer doses than other vaccine platforms as well as eliminating the need for adjuvants. However, virus instability in liquid formulations may limit their shelf life and require that these products be transported and stored under stringently controlled temperature conditions, contributing to high cost and limiting patient access. In this work, spray-drying and lyophilization were used to embed an infectious enveloped virus within dry, glassy polysaccharide matrices. No loss of viral titer was observed following either spray-drying (at multiple drying gas temperatures) or lyophilization. Furthermore, viruses embedded in the glassy formulations showed enhanced thermal stability, retaining infectivity after exposure to elevated temperatures as high as 85 °C for up to one hour, and for up to 10 weeks at temperatures as high as 30 °C. In comparison, viruses in liquid formulations lost infectivity within an hour at temperatures above 40 °C, or after incubation at 25 °C for longer periods of time.

Stain-Free Approach to Determine and Monitor Cell Heath Using Supervised and Unsupervised Image-Based Deep Learning.

Cell-based medicinal products (CBMPs) are a growing class of therapeutics that promise new treatments for complex and rare diseases. Given the inherent complexity of the whole human cells comprising CBMPs, there is a need for robust and fast analytical methods for characterization, process monitoring, and quality control (QC) testing during their manufacture. Existing techniques to evaluate and monitor cell quality typically constitute labor-intensive, expensive, and highly specific staining assays. In this work, we combine image-based deep learning with flow imaging microscopy (FIM) to predict cell health metrics using cellular morphology "fingerprints" extracted from images of unstained Jurkat cells (immortalized human T-lymphocyte cells). A supervised (i.e., algorithm trained with human-generated labels for images) fingerprinting algorithm, trained on images of unstained healthy and dead cells, provides a robust stain-free, non-invasive, and non-destructive method for determining cell viability. Results from the stain-free method are in good agreement with traditional stain-based cytometric viability measurements. Additionally, when trained with images of healthy cells, dead cells and cells undergoing chemically induced apoptosis, the supervised fingerprinting algorithm is able to distinguish between the three cell states, and the results are independent of specific treatments or signaling pathways. We then show that an unsupervised variational autoencoder (VAE) algorithm trained on the same images, but without human-generated labels, is able to distinguish between samples of healthy, dead and apoptotic cells along with cellular debris based on learned morphological features and without human input. With this, we demonstrate that VAEs are a powerful exploratory technique that can be used as a process monitoring analytical tool.