Machine Learning for Deconvolution and Segmentation of Hyperspectral Imaging Data from Biopharmaceutical Resins.

Biopharmaceutical resins are pivotal inert matrices used across industry and academia, playing crucial roles in a myriad of applications. For biopharmaceutical process research and development applications, a deep understanding of the physical and chemical properties of the resin itself is frequently required, including for drug purification, drug delivery, and immobilized biocatalysis. Nevertheless, the prevailing methodologies currently employed for elucidating these important aspects of biopharmaceutical resins are often lacking, frequently require significant sample alteration, are destructive or ionizing in nature, and may not adequately provide representative information. In this work, we propose the use of unsupervised machine learning technologies, in the form of both non-negative matrix factorization (NMF) and k-means segmentation, in conjugation with Raman hyperspectral imaging to rapidly elucidate the molecular and spatial properties of biopharmaceutical resins. Leveraging our proposed technology, we offer a new approach to comprehensively understanding important resin-based systems for application across biopharmaceuticals and beyond. Specifically, focusing herein on a representative resin widely utilized across the industry (i.e., Immobead 150P), our findings showcase the ability of our machine learning-based technology to molecularly identify and spatially resolve all chemical species present. Further, we offer a comprehensive evaluation of optimal excitation for hyperspectral imaging data collection, demonstrating results across 532, 638, and 785 nm excitation. In all cases, our proposed technology deconvoluted, both spatially and spectrally, resin and glass substrates via NMF. After NMF deconvolution, image segmentation was also successfully accomplished in all data sets via k-means clustering. To the best of our knowledge, this is the first report utilizing the combination of two unsupervised machine learning methodologies, combining NMF and k-means, for the rapid deconvolution and segmentation of biopharmaceutical resins. As such, we offer a powerful new data-rich experimentation tool for application across multidisciplinary fields for a deeper understanding of resins.

Occurrence, Concentration, and Distribution of 35 PFASs and Their Precursors Retained in 20 Stormwater Biofilters.

Current knowledge about the fate and transport behaviors of per- and polyfluoroalkyl substances (PFASs) in urban stormwater biofilter facilities is very limited. C5-14,16 perfluoroalkyl carboxylic acids [perfluorinated carboxylic acids (PFCAs)], C4,8,10 perfluoroalkanesulfonic acids (PFSAs), methyl-perfluorooctane sulfonamide acetic acid (MeFOSAA, a PFSA precursor), and unknown C6-8 PFCA and perfluorooctanesulfonic acid precursors were frequently found in bioretention media and forebay sediments at Σ35PFAS concentrations of <0.03-19 and 0.064-16 µg/kg-DW, respectively. Unknown C6-8 PFCA precursor concentrations were up to ten times higher than the corresponding PFCAs, especially at forebays and biofilters' top layer. No significant trend could be attributed to PFAS and precursor concentrations versus depth of filter media, though PFAS concentrations were 2-3 times higher in the upper layers on average (significant difference between the upper (0-5 cm) and deepest (35-50 cm) layer). PFASs had a similar spatial concentration distribution in each filter media (no clear difference between short- and long-chain PFASs). Commercial land use and organic matter were important factors explaining the concentration variations among the biofilters and between the sampling depths, respectively. Given the comparable PFAS accumulations in deeper and superficial layers and possible increased mobility after precursor biotransformation, designing shallow-depth, nonamended sand biofilters or maintaining only the top layer may be insufficient for stormwater PFAS management.

Mosquito abundance and diversity in central Ohio, USA vary among stormwater wetlands, retention ponds, and detention ponds and their associated environmental parameters.

Mosquitoes (Diptera: Culicidae) are one of the most impactful pests to human society, both as a nuisance and a potential vector of human and animal pathogens. Mosquito larvae develop in still aquatic environments. Eliminating these habitats near high human density or managing them to reduce the suitability for mosquitoes will reduce mosquito populations in these human environments and decrease the overall negative impact of mosquitoes on humans. One common source of standing water in urban and suburban environments is the water that pools in stormwater control measures. Previous studies have shown that some stormwater control measures generate large numbers of mosquitoes while others harbor none, and the reason for this difference remains unclear. Our study focuses on elucidating the factors that cause a stormwater control measure to be more or less suitable for mosquitoes. During the summers of 2021 and 2022, we collected and identified mosquito larvae from thirty stormwater control measures across central Ohio to assess variation in mosquito abundance and diversity among sites. Our goal was to determine if specific types of stormwater control measures (retention ponds, detention ponds, or constructed wetlands) harbored different abundances of mosquitoes or different community structures. We also assessed environmental parameters of these sites to elucidate their effects on mosquito abundance and diversity. Overall, we recorded the highest number of mosquito larvae and species in constructed wetlands. However, these sites were dominated by the innocuous species, Culex territans. Conversely, detention ponds held fewer mosquitoes but a higher proportion of known vector species, including Culex pipiens and Aedes vexans. The total number of mosquitoes across all sites was correlated with higher vegetation, more shade, lower water temperatures, and lower pH, suggesting stormwater control measures with these features may also be hotspots for mosquito proliferation.

Plasma from older children in Malawi inhibits Plasmodium falciparum binding in 3D brain microvessels.

A hallmark of cerebral malaria is sequestration of Plasmodium falciparum-infected erythrocytes (IEs) in the brain microcirculation. Antibodies contribute to malaria immunity, but it remains unclear whether functional antibodies targeting parasite-expressed ligand can block cytoadhesion in the brain. Here, we screened the plasma of older children and young adults in Malawi to characterize the antibody response against the P. falciparum-IE surface and used a bioengineered 3D human brain microvessel model incorporating variable flow dynamics to measure adhesion blocking responses. We found a strong correlation between surface antibody reactivity by flow cytometry and reduced P. falciparum-IE binding in 3D microvessels. Moreover, there was a threshold of surface antibody reactivity necessary to achieve robust inhibitory activity. Our findings provide evidence of the acquisition of adhesion blocking antibodies against cerebral binding variants in people exposed to stable P. falciparum transmission and suggest the quality of the inhibitory response can be influenced by flow dynamics.

Analyzing atomic force microscopy images of virus-like particles by expectation-maximization.

Analysis of virus-like particles (VLPs) is an essential task in optimizing their implementation as vaccine antigens for virus-initiated diseases. Interrogating VLP collections for elasticity by probing with a rigid atomic force microscopy (AFM) tip is a potential method for determining VLP morphological changes. During VLP morphological change, it is not expected that all VLPs would be in the same state. This leads to the open question of whether VLPs may change in a continuous or stepwise fashion. For continuous change, the statistical distribution of observed VLP properties would be expected as a single distribution, while stepwise change would lead to a multimodal distribution of properties. This study presents the application of a Gaussian mixture model (GMM), fit by the Expectation-Maximization (EM) algorithm, to identify different states of VLP morphological change observed by AFM imaging.

Chinook salmon depth distributions on the continental shelf are shaped by interactions between location, season, and individual condition.

BACKGROUND: Ecological and physical conditions vary with depth in aquatic ecosystems, resulting in gradients of habitat suitability. Although variation in vertical distributions among individuals provides evidence of habitat selection, it has been challenging to disentangle how processes at multiple spatio-temporal scales shape behaviour. METHODS: We collected thousands of observations of depth from > 300 acoustically tagged adult Chinook salmon Oncorhynchus tshawytscha, spanning multiple seasons and years. We used these data to parameterize a machine-learning model to disentangle the influence of spatial, temporal, and dynamic oceanographic variables while accounting for differences in individual condition and maturation stage. RESULTS: The top performing machine learning model used bathymetric depth ratio (i.e., individual depth relative to seafloor depth) as a response. We found that bathymetry, season, maturation stage, and spatial location most strongly influenced Chinook salmon depth. Chinook salmon bathymetric depth ratios were deepest in shallow water, during winter, and for immature individuals. We also identified non-linear interactions among covariates, resulting in spatially-varying effects of zooplankton concentration, lunar cycle, temperature and oxygen concentration. CONCLUSIONS: Our results suggest Chinook salmon vertical habitat use is a function of ecological interactions, not physiological constraints. Temporal and spatial variation in depth distributions could be used to guide management decisions intended to reduce fishery impacts on Chinook salmon. More generally, our findings demonstrate how complex interactions among bathymetry, seasonality, location, and life history stage regulate vertical habitat selection.

Selective Duplex Formation in Mixed Sequence Libraries of Synthetic Polymers.

Recognition-encoded melamine oligomers (REMO) are synthetic polymers that feature an alternating 1,3,5-triazine-piperazine backbone and side-chains equipped with either a phenol or phosphine oxide recognition unit. An automated method for the solid-phase synthesis (SPS) of REMO of any specified sequence has been developed starting from dichlorotriazine monomer building blocks. Complementary homo-oligomers with either six phenols or six phosphine oxides were synthesized and shown to form a stable duplex in nonpolar solvents by NMR denaturation experiments. The duplex was covalently trapped by equipping the ends of the oligomers with an azide and an alkyne group and using a copper-catalyzed alkyne-azide cycloaddition (CuAAC) reaction. The SPS methodology was adapted to synthesize mixed sequence libraries by using a mixture of two different dichlorotriazine building blocks in each coupling cycle of an oligomer synthesis. The resulting libraries contain statistical mixtures of all possible sequences. The self-assembly properties of these libraries were screened by using the CuAAC reaction to trap any duplexes present. In mixed sequence libraries of 6-mers, the trapping experiments showed that only sequence-complementary oligomers formed duplexes at micromolar concentrations in dichloromethane. The automated synthesis approach developed here provides access to large libraries of mixed sequence synthetic polymers, and the covalent trapping experiment provides a convenient tool for screening functional properties of mixtures. The results suggest high-fidelity sequence-selective duplex formation in mixtures of 6-mer sequences of the REMO architecture.

Interrogating endothelial barrier regulation by temporally resolved kinase network generation.

Kinases are key players in endothelial barrier regulation, yet their temporal function and regulatory phosphosignaling networks are incompletely understood. We developed a novel methodology, Temporally REsolved KInase Network Generation (TREKING), which combines a 28-kinase inhibitor screen with machine learning and network reconstruction to build time-resolved, functional phosphosignaling networks. We demonstrated the utility of TREKING for identifying pathways mediating barrier integrity after activation by thrombin with or without TNF preconditioning in brain endothelial cells. TREKING predicted over 100 kinases involved in barrier regulation and discerned complex condition-specific pathways. For instance, the MAPK-activated protein kinase 2 (MAPKAPK2/MK2) had early barrier-weakening activity in both inflammatory conditions but late barrier-strengthening activity exclusively with thrombin alone. Using temporal Western blotting, we confirmed that MAPKAPK2/MK2 was differentially phosphorylated under the two inflammatory conditions. We further showed with lentivirus-mediated knockdown of MAPK14/p38α and drug targeting the MAPK14/p38α-MAPKAPK2/MK2 complex that a MAP3K20/ZAK-MAPK14/p38α axis controlled the late activation of MAPKAPK2/MK2 in the thrombin-alone condition. Beyond the MAPKAPK2/MK2 switch, TREKING predicts extensive interconnected networks that control endothelial barrier dynamics.

The potential to manage releases of Bacillus anthracis using bioretention and a high flow media filter: Results of simulated runoff testing with tracer spores Bacillus globigii.

The threat of bioterrorism has spurred research on the decontamination and containment of different agents. Anthrax [causative agent Bacillus anthracis (Ba)] is a disease that can lead to severe infections within human and animals, particularly when inhaled. This research investigated the use of spore-contaminated simulated runoff events into stormwater control measures (SCMs), which are designed to retain and improve the quality of runoff and may have the potential to filter and contain the spores. In this study, the effectiveness of a bioretention cell (BRC) and high flow media filter (HFMF) in Huron, Ohio, were evaluated for removal of Bacillus globigii (Bg) spores (a harmless cognate of Ba). Three 4-8 mm simulated runoff events were created for each SCM using a fire hydrant and Bg spores were injected into the runoff upstream of the SCM inlets. The BRC significantly (p < 0.001) outperformed the HFMF in reducing Bg concentrations and loads, with an average load reduction of 1.9 log (∼99% reduction) compared to 0.4 (∼60% reduction), respectively. A probable critical design factor leading to these differences was the infiltration rate of the media and subsequent retention time within the filters, which was supported by similar disparities in suspended solids reductions. Differences in spore removal may also have been due to particle size distribution of the HFMF, which was more gravelly than the bioretention cell. At 3 and 6 months after the-simulated runoff tests, soil samples taken from both SCMs, yielding detectable Bg spores within the top 15 cm of media, with increased spore concentrations where ponding occurred for longer durations during the tests. This suggests that forebays and areas near inlets may be hotspots for spore cleanup in a real-world bioterrorism incident.

Unique Interactions of the Small Translocases of the Mitochondrial Inner Membrane (Tims) in Trypanosoma brucei.

The infectious agent for African trypanosomiasis, Trypanosoma brucei, possesses a unique and essential translocase of the mitochondrial inner membrane, known as the TbTIM17 complex. TbTim17 associates with six small TbTims (TbTim9, TbTim10, TbTim11, TbTim12, TbTim13, and TbTim8/13). However, the interaction patterns of these smaller TbTims with each other and TbTim17 are not clear. Through yeast two-hybrid (Y2H) and co-immunoprecipitation analyses, we demonstrate that all six small TbTims interact with each other. Stronger interactions were found among TbTim8/13, TbTim9, and TbTim10. However, TbTim10 shows weaker associations with TbTim13, which has a stronger connection with TbTim17. Each of the small TbTims also interacts strongly with the C-terminal region of TbTim17. RNAi studies indicated that among all small TbTims, TbTim13 is most crucial for maintaining the steady-state levels of the TbTIM17 complex. Further analysis of the small TbTim complexes by size exclusion chromatography revealed that each small TbTim, except for TbTim13, is present in ~70 kDa complexes, possibly existing in heterohexameric forms. In contrast, TbTim13 is primarily present in the larger complex (>800 kDa) and co-fractionates with TbTim17. Altogether, our results demonstrate that, relative to other eukaryotes, the architecture and function of the small TbTim complexes are specific to T. brucei.

Comparison of Patient Satisfaction Between Face-to-Face and Telehealth Modalities in Nephrolithiasis Nutritional Counseling.

INTRODUCTION: Following the COVID-19 pandemic, telehealth usage increased. Virtual visits minimize exposure risk while also addressing barriers to care. Telehealth offers the ability to increase patient access and provider efficiency. However, patient satisfaction with telehealth has not been fully determined. This study evaluated patient-perceived quality and satisfaction of virtual vs face-to-face visits during consultation with a dietician in the management of nephrolithiasis. METHODS: Ninety-six patients with previous diagnosis of nephrolithiasis underwent an initial, in-person nutrition consultation between May 2019 to February 2021. A follow-up with a dietician was randomized to in-person or telehealth. The telehealth group used an application called MDLive. The telehealth group used MDLive on a hospital computer during their follow-up with the urologist, whereas the in-person group had a separate appointment scheduled at a different location. Patient satisfaction following telehealth visits was assessed by the Telemedicine Satisfaction Questionnaire. Patient satisfaction following in-person visits was assessed with an 8-question modified Telemedicine Satisfaction Questionnaire, which lacked technology-related questions. RESULTS: Fifty patients were randomized to in-person follow-up and 46 to virtual follow-up. Within the virtual follow-up group more than 90% "agreed" or "strongly agreed" that they were satisfied with the quality of service provided through telemedicine. Greater than 82% reported intention to use telemedicine services again. There was no significant difference in patient satisfaction between telemedicine and face-to-face visits. Sixty-seven percent of patients in the telemedicine group reported better access to health care services and time saved and 89% reported independence accessing the telehealth system without assistance. CONCLUSIONS: This study supports the idea that telemedicine may be a successful alternative in the follow up of patients undergoing nutritional counseling for stone prevention. Future studies regarding telehealth use should evaluate which other urologic conditions are amenable to virtual management.

A genetically-encoded fluorescent biosensor for visualization of acetyl-CoA in live cells.

Acetyl-coenzyme A is a central metabolite that participates in many cellular pathways. Evidence suggests that acetyl-CoA production and consumption are highly compartmentalized in mammalian cells. Yet methods to measure acetyl-CoA in living cells are lacking. In this work, we engineer an acetyl-CoA biosensor from the bacterial protein PanZ and circularly permuted green fluorescent protein (cpGFP). We biochemically characterize the sensor and demonstrate its selectivity for acetyl-CoA over other CoA species. We then deploy the biosensor in E. coli and HeLa cells to demonstrate its utility in living cells. In E. coli, we show that the biosensor enables detection of rapid changes in acetyl-CoA levels. In human cells, we show that the biosensor enables subcellular detection and reveals the compartmentalization of acetyl-CoA metabolism.

In situ process analytical technology for real time viable cell density and cell viability during live-virus vaccine production.

Viable cell density (VCD) and cell viability (CV) are key performance indicators of cell culture processes in biopharmaceutical production of biologics and vaccines. Traditional methods for monitoring VCD and CV involve offline cell counting assays that are both labor intensive and prone to high variability, resulting in sparse sampling and uncertainty in the obtained data. Process analytical technology (PAT) approaches offer a means to address these challenges. Specifically, in situ probe-based measurements of dielectric spectroscopy (also commonly known as capacitance) can characterize VCD and CV continuously in real time throughout an entire process, enabling robust process characterization. In this work, we propose in situ dielectric spectroscopy as a PAT tool for real time analysis of live-virus vaccine (LVV) production. Dielectric spectroscopy was collected across 25 discreet frequencies, offering a thorough evaluation of the proposed technology. Correlation of this PAT methodology to traditional offline cell counting assays was performed, in which VCD and CV were both successfully predicted using dielectric spectroscopy. Both univariate and multivariate data analysis approaches were evaluated for their potential to establish correlation between the in situ dielectric spectroscopy and offline measurements. Univariate analysis strategies are presented for optimal single frequency selection. Multivariate analysis, in the form of partial least squares (PLS) regression, produced significantly higher correlations between dielectric spectroscopy and offline VCD and CV data, as compared to univariate analysis. Specifically, by leveraging multivariate analysis of dielectric information from all 25 spectroscopic frequencies measured, PLS models performed significantly better than univariate models. This is particularly evident during cell death, where tracking VCD and CV have historically presented the greatest challenge. The results of this work demonstrate the potential of both single and multiple frequency dielectric spectroscopy measurements for enabling robust LVV process characterization, suggesting that broader application of in situ dielectric spectroscopy as a PAT tool in LVV processes can provide significantly improved process understanding. To the best of our knowledge, this is the first report of in situ dielectric spectroscopy with multivariate analysis to successfully predict VCD and CV in real time during live virus-based vaccine production.

"A friendly reminder" - Improving workflow and efficiency in a pulmonary fellows' outpatient continuity clinic.

BACKGROUND: Seeing patients in an ambulatory clinic generates electronic medical record (EMR) inbox tasks. Little is known about the standard baseline message turnaround time to EMR inbox task completion and whether electronic reminders improve turnaround time. OBJECTIVE: 1) Obtain baseline message type and mean message turnaround time (MTT) to EMR inbox task completion data, 2) Standardize EMR workflow education, 3) Disseminate bi-weekly electronic reminders to fellows in their continuity clinic and measure MTT. METHODS: Prospective, non-randomized, unblinded, cross-over pre- and post-intervention pilot study in an ambulatory pulmonary clinic at a large, urban, academic referral health system. Sixteen pulmonary and critical care fellows affiliated with the Indiana University School of Medicine Pulmonary and Critical Care Fellowship were divided equally into two groups, with the study period from October of 2021 to May of 2022, and were given bi-weekly calendar reminders in Microsoft Outlook with measurement of EMR messages and MTT. RESULTS: 2554 messages were acknowledged with result notes (n = 1676, 59.16 %) being the most common. There was a 40 % decrease in overall MTT from the pre- to the post-intervention period (MTT = 33 days in pre-intervention period for whole cohort, MTT = 19 days in post-intervention period). CONCLUSIONS: MTT for EMR inbox tasks at a large, academic center with fellowship trainees is roughly 2.5 weeks. These findings should prompt other institutions to investigate their own trainees' inbox handling habits and validates the benefit of EMR training and reminders on fellowship trainee's in-basket task turnaround time.

Exacerbations, treatment patterns, utilization, and costs before and after initiating of benralizumab for the treatment of severe eosinophilic asthma.

OBJECTIVES: The purpose of this study was to examine the number of exacerbations, counts of eosinophils, and asthma-related symptoms 1 year before and after initiating benralizumab for the treatment of severe eosinophilic asthma. METHODS: Patients with prior exacerbations and newly initiating benralizumab were identified in the claims-based Healthcare Integrated Research Database. Claims were used to assess benralizumab treatment patterns, exacerbations, healthcare resource utilization, and other asthma medication used. Among a subset of patients, medical records were abstracted for Asthma Control Test (ACT) scores and asthma symptoms. RESULTS: There were 506 patients meeting inclusion/exclusion criteria for claims-based analyses and 123 for medical-record analyses. The number of patients experiencing exacerbations significantly decreased from baseline to follow-up (40% reduction, McNemar's χ2 = 204.00, p < .001). The mean number of exacerbations also decreased from 3.2 (1.5) to 1.2 (1.4) (paired t = 24.45, p < .001; Cohen's D = 1.09). The effects were larger among patients with eosinophils ≥300 cells/µL. Among patients with an ACT available for baseline and follow-up (n = 47), there was a significant reduction in the number of patients with scores <19 (72% vs. 45%, p < .01). CONCLUSIONS: Treatment with benralizumab resulted in fewer exacerbations, reduced utilization, and improved ACT scores. This study demonstrates that benralizumab is an effective treatment option for patients with severe eosinophilic asthma.

Muscle specific declines in oxygen saturation during acute ambulation with hands-free and conventional mobility devices.

Disuse is associated with reduced muscle oxygen saturation (SmO2). Improving oxygen delivery to tissues is important for healing, preventing muscle atrophy, and reducing the risk of deep vein thrombosis. Mobility devices are used during disuse periods to ambulate and protect the injured limb. This study examined SmO2 in walking and ambulation with various mobility devices. Thirty-eight participants randomly completed four, ten-minute trials which included: (1) walking, (2) medical kneeling scooter (MKS), (3) hands-free crutch (HFC), and (4) axillary crutch (AC). During each trial, near infrared spectroscopy sensors were placed on the vastus lateralis (VL), biceps femoris (BF), and lateral gastrocnemius (LG) of the right limb. Compared to walking, all mobility devices showed a decline in SmO2 in the VL of ∼10% (mean ± SD; 75% ± 12%-65% ± 17%, P < 0.05). In the BF, SmO2 declined ∼9% in AC compared to walking (76% ± 12%-67% ± 17%, P = 0.025). In the LG, SmO2 declined in AC (64% ± 16%) compared to MKS (70% ± 15%, P = 0.005). There were no differences in LG SmO2 compared to walking (69% ± 13%) in MKS (P > 0.05) or HFC (65% ± 15%, P > 0.05). In young, healthy volunteers, the use of mobility devices altered muscle oxygenation in several muscles. AC reduced muscle oxygenation in the VL, BF, and LG; while MKS and HFC maintained BF and LG muscle oxygenation at a level consistent with ambulatory walking.

Probing cerebral malaria inflammation in 3D human brain microvessels.

Sequestration of Plasmodium falciparum-infected erythrocytes (IEs) in the brain microcirculation is a hallmark of cerebral malaria (CM), which leads to endothelial activation, brain swelling, and death. Here, we probed CM inflammation in a perfusable 3D human brain microvessel model. 3D brain microvessels supported in vivo-like capacities for parasite binding and maturation in situ, leading to a distinct inflammatory response from the pro-inflammatory cytokine tumor necrosis factor α (TNF-α). By combining transcriptional analysis, imaging, and leukocyte perfusion, we showed that whereas TNF-α promotes a reversible inflammatory phenotype with widespread leukocyte recruitment, parasites induce unique stress response pathways and cause localized cell adhesivity changes, focal endothelial disruptions, and apoptosis. Furthermore, parasites modified the temporal kinetics of the TNF transcriptional response, suggesting augmented inflammatory damage with the two sequential stimuli. Our findings offer mechanistic insights into CM biology in a 3D brain microvessel mimetic platform and suggest that multiple events intersect to promote brain barrier inflammation in CM.

Modernized Machine Learning Approach to Illuminate Enzyme Immobilization for Biocatalysis.

Biocatalysis is an established technology with significant application in the pharmaceutical industry. Immobilization of enzymes offers significant benefits for commercial and practical purposes to enhance the stability and recyclability of biocatalysts. Determination of the spatial and chemical distributions of immobilized enzymes on solid support materials is essential for an optimal catalytic performance. However, current analytical methodologies often fall short of rapidly identifying and characterizing immobilized enzyme systems. Herein, we present a new analytical methodology that combines non-negative matrix factorization (NMF)-an unsupervised machine learning tool-with Raman hyperspectral imaging to simultaneously resolve the spatial and spectral characteristics of all individual species involved in enzyme immobilization. Our novel approach facilitates the determination of the optimal NMF model using new data-driven, quantitative selection criteria that fully resolve all chemical species present, offering a robust methodology for analyzing immobilized enzymes. Specifically, we demonstrate the ability of NMF with Raman hyperspectral imaging to resolve the spatial and spectral profiles of an engineered pantothenate kinase immobilized on two different commercial microporous resins. Our results demonstrate that this approach can accurately identify and spatially resolve all species within this enzyme immobilization process. To the best of our knowledge, this is the first report of NMF within hyperspectral imaging for enzyme immobilization analysis, and as such, our methodology can now provide a new powerful tool to streamline biocatalytic process development within the pharmaceutical industry.