Differentiation between descending thoracic aortic diseases using machine learning and plasma proteomic signatures.

BACKGROUND: Descending thoracic aortic aneurysms and dissections can go undetected until severe and catastrophic, and few clinical indices exist to screen for aneurysms or predict risk of dissection. METHODS: This study generated a plasma proteomic dataset from 75 patients with descending type B dissection (Type B) and 62 patients with descending thoracic aortic aneurysm (DTAA). Standard statistical approaches were compared to supervised machine learning (ML) algorithms to distinguish Type B from DTAA cases. Quantitatively similar proteins were clustered based on linkage distance from hierarchical clustering and ML models were trained with uncorrelated protein lists across various linkage distances with hyperparameter optimization using fivefold cross validation. Permutation importance (PI) was used for ranking the most important predictor proteins of ML classification between disease states and the proteins among the top 10 PI protein groups were submitted for pathway analysis. RESULTS: Of the 1,549 peptides and 198 proteins used in this study, no peptides and only one protein, hemopexin (HPX), were significantly different at an adjusted p < 0.01 between Type B and DTAA cases. The highest performing model on the training set (Support Vector Classifier) and its corresponding linkage distance (0.5) were used for evaluation of the test set, yielding a precision-recall area under the curve of 0.7 to classify between Type B from DTAA cases. The five proteins with the highest PI scores were immunoglobulin heavy variable 6-1 (IGHV6-1), lecithin-cholesterol acyltransferase (LCAT), coagulation factor 12 (F12), HPX, and immunoglobulin heavy variable 4-4 (IGHV4-4). All proteins from the top 10 most important groups generated the following significantly enriched pathways in the plasma of Type B versus DTAA patients: complement activation, humoral immune response, and blood coagulation. CONCLUSIONS: We conclude that ML may be useful in differentiating the plasma proteome of highly similar disease states that would otherwise not be distinguishable using statistics, and, in such cases, ML may enable prioritizing important proteins for model prediction.

Parallelization with Dual-Trap Single-Column Configuration Maximizes Throughput of Proteomic Analysis.

Proteomic analysis on the scale that captures population and biological heterogeneity over hundreds to thousands of samples requires rapid mass spectrometry methods, which maximize instrument utilization (IU) and proteome coverage while maintaining precise and reproducible quantification. To achieve this, a short liquid chromatography gradient paired to rapid mass spectrometry data acquisition can be used to reproducibly quantify a moderate set of analytes. High-throughput profiling at a limited depth is becoming an increasingly utilized strategy for tackling large sample sets but the time spent on loading the sample, flushing the column(s), and re-equilibrating the system reduces the ratio of meaningful data acquired to total operation time and IU. The dual-trap single-column configuration (DTSC) presented here maximizes IU in rapid analysis (15 min per sample) of blood and cell lysates by parallelizing trap column cleaning and sample loading and desalting with the analysis of the previous sample. We achieved 90% IU in low microflow (9.5 µL/min) analysis of blood while reproducibly quantifying 300-400 proteins and over 6000 precursor ions. The same IU was achieved for cell lysates and over 4000 proteins (3000 at CV below 20%) and 40,000 precursor ions were quantified at a rate of 15 min/sample. Thus, DTSC enables high-throughput epidemiological blood-based biomarker cohort studies and cell-based perturbation screening.

Patterns of Prescribing Antiretroviral Therapy Upon Discharge to Psychiatry Inpatients With HIV/AIDS at a Large Urban Hospital.

BACKGROUND: HIV infection is more prevalent among people with severe mental illness (SMI) than in the general population. People with SMI may lack access to recommended antiretroviral therapy (ART), and inpatient psychiatric admissions may be opportunities to ensure that individuals receive recommended treatment. OBJECTIVE: To evaluate ART prescription patterns on an inpatient psychiatry service. METHODS: In this retrospective, observational study, patient and admission characteristics and ART prescriptions were obtained for 248 HIV-positive inpatients between 2006 and 2012. Receipt of any ART, any recommended ART regimen, and ART with potentially harmful adverse events and drug interactions were examined. General estimating equation models were used to evaluate prescription patterns in relation to patient and admission characteristics. RESULTS: ART was prescribed at 39% of discharges and increased by 51% during the study. Prescription was more common in admissions with an AIDS diagnosis and age greater than 29 years and less common in admissions associated with a psychotic diagnosis and shorter inpatient stays. When ART was prescribed, regimens were consistent with guideline recommendations 91% of the time. Prescription of potentially harmful regimens was limited. CONCLUSION AND RELEVANCE: In an acute inpatient psychiatry setting in an urban HIV/AIDS epicenter, where psychotic disorders and brief and involuntary admissions were the norm, guideline-recommended ART regimens were prescribed at almost 60% of discharges by the end of the study. Future studies should explore interventions to increase ART for high-risk subpopulations with SMI, including younger individuals or those with brief inpatient psychiatry hospitalizations.

Increasing HIV Testing in Inpatient Psychiatry.

BACKGROUND: People with serious mental illness (SMI) are at elevated risk of HIV infection, but do not receive HIV tests regularly. Inpatient psychiatric admissions provide opportunities for HIV testing. OBJECTIVE: This study retrospectively examined the impact of three sequential interventions designed to increase HIV testing on an acute inpatient psychiatry service: (1) advocacy by an administrative champion, (2) an on-site HIV counselor, and (3) a clinician championing HIV testing. METHOD: Demographic and HIV testing data were extracted from hospital data systems for 11,360 admissions of HIV-negative patients to an inpatient psychiatry service between 2006 and 2012. Relationships among interventions, length of stay, patient demographics, and receipt of an HIV test were examined using general estimating equation methods. RESULTS: In the year prior to the intervention, 7.2% of psychiatric inpatients received HIV tests. After 1 year of administrative advocacy, 11.2% received tests. Following the HIV counseling intervention, 25.1% of patients were tested. After the counseling intervention ended, continued administrative and clinical advocacy was associated with further increases in testing. In the final year studied, 30.3% of patients received HIV tests. Patients with shorter inpatient stays and those of Black or Asian race/ethnicity were less likely to be tested. Further, 1.6% of HIV tests were positive. CONCLUSION: Three interventions of varying intensity were associated with a 5-fold increase in HIV testing on an acute inpatient psychiatry service. Nonetheless, 70% of inpatients were not tested. Continued efforts are needed to increase HIV testing in inpatient psychiatric settings.

Simultaneous Multi-Omics Analysis by Direct Infusion Mass Spectrometry (SMAD-MS).

Combined multi-omics analysis of proteomics, polar metabolomics, and lipidomics requires separate liquid chromatography-mass spectrometry (LC-MS) platforms for each omics layer. This requirement for different platforms limits throughput and increases costs, preventing the application of mass spectrometry-based multi-omics to large scale drug discovery or clinical cohorts. Here, we present an innovative strategy for simultaneous multi-omics analysis by direct infusion (SMAD) using one single injection without liquid chromatography. SMAD allows quantification of over 9,000 metabolite m/z features and over 1,300 proteins from the same sample in less than five minutes. We validated the efficiency and reliability of this method and then present two practical applications: mouse macrophage M1/M2 polarization and high throughput drug screening in human 293T cells. Finally, we demonstrate relationships between proteomic and metabolomic data are discovered by machine learning.

Complete Workflow for High Throughput Human Single Skeletal Muscle Fiber Proteomics.

Skeletal muscle is a major regulatory tissue of whole-body metabolism and is composed of a diverse mixture of cell (fiber) types. Aging and several diseases differentially affect the various fiber types, and therefore, investigating the changes in the proteome in a fiber-type specific manner is essential. Recent breakthroughs in isolated single muscle fiber proteomics have started to reveal heterogeneity among fibers. However, existing procedures are slow and laborious requiring two hours of mass spectrometry time per single muscle fiber; 50 fibers would take approximately four days to analyze. Thus, to capture the high variability in fibers both within and between individuals requires advancements in high throughput single muscle fiber proteomics. Here we use a single cell proteomics method to enable quantification of single muscle fiber proteomes in 15 minutes total instrument time. As proof of concept, we present data from 53 isolated skeletal muscle fibers obtained from two healthy individuals analyzed in 13.25 hours. Adapting single cell data analysis techniques to integrate the data, we can reliably separate type 1 and 2A fibers. Sixty-five proteins were statistically different between clusters indicating alteration of proteins involved in fatty acid oxidation, muscle structure and regulation. Our results indicate that this method is significantly faster than prior single fiber methods in both data collection and sample preparation while maintaining sufficient proteome depth. We anticipate this assay will enable future studies of single muscle fibers across hundreds of individuals, which has not been possible previously due to limitations in throughput.

A Complete Workflow for High Throughput Human Single Skeletal Muscle Fiber Proteomics.

Skeletal muscle is a major regulatory tissue of whole-body metabolism and is composed of a diverse mixture of cell (fiber) types. Aging and several diseases differentially affect the various fiber types, and therefore, investigating the changes in the proteome in a fiber-type specific manner is essential. Recent breakthroughs in isolated single muscle fiber proteomics have started to reveal heterogeneity among fibers. However, existing procedures are slow and laborious, requiring 2 h of mass spectrometry time per single muscle fiber; 50 fibers would take approximately 4 days to analyze. Thus, to capture the high variability in fibers both within and between individuals requires advancements in high throughput single muscle fiber proteomics. Here we use a single cell proteomics method to enable quantification of single muscle fiber proteomes in 15 min total instrument time. As proof of concept, we present data from 53 isolated skeletal muscle fibers obtained from two healthy individuals analyzed in 13.25 h. Adapting single cell data analysis techniques to integrate the data, we can reliably separate type 1 and 2A fibers. Ninety-four proteins were statistically different between clusters indicating alteration of proteins involved in fatty acid oxidation, oxidative phosphorylation, and muscle structure and contractile function. Our results indicate that this method is significantly faster than prior single fiber methods in both data collection and sample preparation while maintaining sufficient proteome depth. We anticipate this assay will enable future studies of single muscle fibers across hundreds of individuals, which has not been possible previously due to limitations in throughput.

Differentiation between Descending Thoracic Aortic Diseases using Machine Learning and Plasma Proteomic Signatures.

Background: Descending thoracic aortic aneurysms and dissections can go undetected until severe and catastrophic, and few clinical indices exist to screen for aneurysms or predict risk of dissection. Methods: This study generated a plasma proteomic dataset from 75 patients with descending type B dissection (Type B) and 62 patients with descending thoracic aortic aneurysm (DTAA). Standard statistical approaches were compared to supervised machine learning (ML) algorithms to distinguish Type B from DTAA cases. Quantitatively similar proteins were clustered based on linkage distance from hierarchical clustering and ML models were trained with uncorrelated protein lists across various linkage distances with hyperparameter optimization using 5-fold cross validation. Permutation importance (PI) was used for ranking the most important predictor proteins of ML classification between disease states and the proteins among the top 10 PI protein groups were submitted for pathway analysis. Results: Of the 1,549 peptides and 198 proteins used in this study, no peptides and only one protein, hemopexin (HPX), were significantly different at an adjusted p-value <0.01 between Type B and DTAA cases. The highest performing model on the training set (Support Vector Classifier) and its corresponding linkage distance (0.5) were used for evaluation of the test set, yielding a precision-recall area under the curve of 0.7 to classify between Type B from DTAA cases. The five proteins with the highest PI scores were immunoglobulin heavy variable 6-1 (IGHV6-1), lecithin-cholesterol acyltransferase (LCAT), coagulation factor 12 (F12), HPX, and immunoglobulin heavy variable 4-4 (IGHV4-4). All proteins from the top 10 most important correlated groups generated the following significantly enriched pathways in the plasma of Type B versus DTAA patients: complement activation, humoral immune response, and blood coagulation. Conclusions: We conclude that ML may be useful in differentiating the plasma proteome of highly similar disease states that would otherwise not be distinguishable using statistics, and, in such cases, ML may enable prioritizing important proteins for model prediction.

Comprehensive Overview of Bottom-Up Proteomics using Mass Spectrometry.

Proteomics is the large scale study of protein structure and function from biological systems through protein identification and quantification. "Shotgun proteomics" or "bottom-up proteomics" is the prevailing strategy, in which proteins are hydrolyzed into peptides that are analyzed by mass spectrometry. Proteomics studies can be applied to diverse studies ranging from simple protein identification to studies of proteoforms, protein-protein interactions, protein structural alterations, absolute and relative protein quantification, post-translational modifications, and protein stability. To enable this range of different experiments, there are diverse strategies for proteome analysis. The nuances of how proteomic workflows differ may be challenging to understand for new practitioners. Here, we provide a comprehensive overview of different proteomics methods to aid the novice and experienced researcher. We cover from biochemistry basics and protein extraction to biological interpretation and orthogonal validation. We expect this work to serve as a basic resource for new practitioners in the field of shotgun or bottom-up proteomics.

Bias or biology? Importance of model interpretation in machine learning studies from electronic health records.

Objective: The rate of diabetic complication progression varies across individuals and understanding factors that alter the rate of complication progression may uncover new clinical interventions for personalized diabetes management. Materials and Methods: We explore how various machine learning (ML) models and types of electronic health records (EHRs) can predict fast versus slow onset of neuropathy, nephropathy, ocular disease, or cardiovascular disease using only patient data collected prior to diabetes diagnosis. Results: We find that optimized random forest models performed best to accurately predict the diagnosis of a diabetic complication, with the most effective model distinguishing between fast versus slow nephropathy (AUROC = 0.75). Using all data sets combined allowed for the highest model predictive performance, and social history or laboratory alone were most predictive. SHapley Additive exPlanations (SHAP) model interpretation allowed for exploration of predictors of fast and slow complication diagnosis, including underlying biases present in the EHR. Patients in the fast group had more medical visits, incurring a potential informed decision bias. Discussion: Our study is unique in the realm of ML studies as it leverages SHAP as a starting point to explore patient markers not routinely used in diabetes monitoring. A mix of both bias and biological processes is likely present in influencing a model's ability to distinguish between groups. Conclusion: Overall, model interpretation is a critical step in evaluating validity of a user-intended endpoint for a model when using EHR data, and predictors affected by bias and those driven by biologic processes should be equally recognized.

Glycemic Control in Adult Surgical Patients Receiving Regular Insulin Added to Parenteral Nutrition vs Insulin Glargine: A Retrospective Chart Review.

BACKGROUND: Hyperglycemia is a frequent complication in patients receiving parenteral nutrition (PN) and has been associated with an increased risk of mortality. Treatment of hyperglycemia requires insulin therapy; however, the optimal dose and route have not been established. This study aimed to compare regular insulin added to PN (RI-in-PN) with subcutaneous insulin glargine for the management of hyperglycemia in patients receiving PN. METHODS: This retrospective study was conducted at a tertiary medical center and reviewed 113 adult, non-critically ill surgical patient admissions receiving PN over a 5-year period. The primary outcome was achievement of glycemic control. Secondary outcomes were time to glycemic control, hypoglycemic events, hospital length of stay, and 1-year mortality. RESULTS: The RI-in-PN group had a significantly higher percentage of patient admissions who achieved glycemic control compared with the insulin glargine group (71.8% vs 48.6%, P = 0.017). There was no difference in time to glycemic control, hypoglycemic events, hospital length of stay, or 1-year mortality between groups. Among patients with diabetes mellitus (DM), however, the insulin glargine group had a significantly higher percentage of admissions with at least 1 hypoglycemic event (45.5% vs 20%, P = 0.035). CONCLUSIONS: RI-in-PN is recommended over insulin glargine because of the higher likelihood of achieving glycemic control and, in patients with DM, lower risk of hypoglycemic events. Large, randomized controlled trials are needed to further guide prescribing practice.