Deep Feature Engineering in Colposcopy Image Recognition: A Comparative Study.

Feature fusion techniques have been proposed and tested for many medical applications to improve diagnostic and classification problems. Specifically, cervical cancer classification can be improved by using such techniques. Feature fusion combines information from different datasets into a single dataset. This dataset contains superior discriminant power that can improve classification accuracy. In this paper, we conduct comparisons among six selected feature fusion techniques to provide the best possible classification accuracy of cervical cancer. The considered techniques are canonical correlation analysis, discriminant correlation analysis, least absolute shrinkage and selection operator, independent component analysis, principal component analysis, and concatenation. We generate ten feature datasets that come from the transfer learning of the most popular pre-trained deep learning models: Alex net, Resnet 18, Resnet 50, Resnet 10, Mobilenet, Shufflenet, Xception, Nasnet, Darknet 19, and VGG Net 16. The main contribution of this paper is to combine these models and then apply them to the six feature fusion techniques to discriminate various classes of cervical cancer. The obtained results are then fed into a support vector machine model to classify four cervical cancer classes (i.e., Negative, HISL, LSIL, and SCC). It has been found that the considered six techniques demand relatively comparable computational complexity when they are run on the same machine. However, the canonical correlation analysis has provided the best performance in classification accuracy among the six considered techniques, at 99.7%. The second-best methods were the independent component analysis, least absolute shrinkage and the selection operator, which were found to have a 98.3% accuracy. On the other hand, the worst-performing technique was the principal component analysis technique, which offered 90% accuracy. Our developed approach of analysis can be applied to other medical diagnosis classification problems, which may demand the reduction of feature dimensions as well as a further enhancement of classification performance.

Urine proteomics by mass spectrometry identifies proteins involved in key pathogenic pathways in patients with juvenile dermatomyositis.

OBJECTIVES: To identify and validate biomarkers in JDM patients using a multiplexing tandem mass tag urine proteome profiling approach. METHODS: First morning void urine samples were collected from JDM patients (n = 20) and healthy control subjects (n = 21) and processed for analysis using a standardized liquid chromatography-tandem mass spectrometry approach. Biomarkers with significantly altered levels were correlated with clinical measures of myositis disease activity and damage. A subset of candidate biomarkers was validated using commercially available ELISA kits. RESULTS: In total, 2348 proteins were detected in the samples, with 275 proteins quantified across all samples. Among the differentially altered proteins, cathepsin D and galectin-3 binding protein were significantly increased in the urine of JDM patients (adjusted P < 0.05), supporting previous findings in myositis patients. These two candidate biomarkers were confirmed with ELISAs. Cathepsin D positively correlated with Myositis Damage Index (r = 0.57, P < 0.05) and negatively correlated with the Childhood Myositis Assessment Scale (r = -0.54, P < 0.05). We also identified novel JDM candidate biomarkers involved with key features of myositis, including extracellular matrix remodelling proteins. CONCLUSION: This study confirmed the presence of several proteins in the urine of JDM patients that were previously found to be elevated in the blood of myositis patients and identified novel candidate biomarkers that require validation. These results support the use of urine as a source for biomarker development in JDM.

Acute serum protein and cytokine response of single dose of prednisone in adult volunteers.

Pharmacological glucocorticoids are the most prescribed anti-inflammatory medications, and are chemical variants of cortisol, the circadian and stress hormone. Both endogenous and pharmacological glucocorticoids bind the glucocorticoid receptor (NR3C1) with high affinity, and both then bind downstream gene promoter elements (GRE) to drive positive gene transcription of many proteins. Glucocorticoid/GR complexes also bind distinct negative gene promoter elements (nGRE) to inhibit expression of genes involved in NF-κB innate immunity signaling. We sought to define the acute response of a single dose of prednisone (0.2 mg/kg) in young adult volunteers, with blood samples taken at baseline, 2, 3, 4 and 6 h post-oral dose. To control for circadian morning cortisol hitting the same molecular pathways, a day of blood draws was done without oral prednisone (same time of day), one day prior to drug day. Serum samples were processed for steroid hormone profiles (mass spectrometry; 9 steroidal hormones), proteomics (SOMAscan aptamer panels, 1,305 proteins), and inflammatory markers (Meso Scale Discovery; 10 pro-inflammatory cytokines). The pharmacological effect of the prednisone dose was shown by significant declines of adrenal steroids by 3 h after dosing. IL-10 showed drug-related increase to 4 hrs, then decrease to 6 hrs. IL-8 showed drug-related decrease in serum by 4 h, consistent with direct negative action of GR/ligand on IL-8 gene promoter. Proteomics data showed beta-2 microglobulin, TNFSF15, TSH, CST3, NBL1 to show time-related decreases with prednisone, while CXCL13 showed increases, although these require validation. In summary, a single low dose of prednisone leads to broad suppression of the adrenal axis within 3 h, and down-regulation of inflammatory serum proteins by 6 h.

Temporal Proteomic Profiling During Differentiation of Normal and Dystrophin-Deficient Human Muscle Cells.

BACKGROUND: Myogenesis is a dynamic process involving temporal changes in the expression of many genes. Lack of dystrophin protein such as in Duchenne muscular dystrophy might alter the natural course of gene expression dynamics during myogenesis. OBJECTIVE: To gain insight into the dynamic temporal changes in protein expression during differentiation of normal and dystrophin deficient myoblasts to myotubes. METHOD: A super SILAC spike-in strategy in combination and LC-MS/MS was used for temporal proteome profiling of normal and dystrophin deficient myoblasts during differentiation. The acquired data was analyzed using Proteome Discoverer 2.2. and data clustering using R to define significant temporal changes in protein expression. RESULTS: sFour major temporal protein clusters that showed sequential dynamic expression profiles during myogenesis of normal myoblasts were identified. Clusters 1 and 2, consisting mainly of proteins involved mRNA splicing and processing expression, were elevated at days 0 and 0.5 of differentiation then gradually decreased by day 7 of differentiation, then remained lower thereafter. Cluster 3 consisted of proteins involved contractile muscle and actomyosin organization. They increased in their expression reaching maximum at day 7 of differentiation then stabilized thereafter. Cluster 4 consisting of proteins involved in skeletal muscle development glucogenesis and extracellular remodeling had a lower expression during myoblast stage then gradually increased in their expression to reach a maximum at days 11-15 of differentiation. Lack of dystrophin expression in DMD muscle myoblast caused major alteration in temporal expression of proteins involved in cell adhesion, cytoskeleton, and organelle organization as well as the ubiquitination machinery. CONCLUSION: Time series proteome profiling using super SILAC strategy is a powerful method to assess temporal changes in protein expression during myogenesis and to define the downstream consequences of lack of dystrophin on these temporal protein expressions. Key alterations were identified in dystrophin deficient myoblast differentiation compared to normal myoblasts. These alterations could be an attractive therapeutic target.

Serum protein biomarkers for juvenile dermatomyositis: a pilot study.

BACKGROUND: Blood accessible biomarkers to assess disease activity and their response to therapies in Juvenile Dermatomyositis (JDM) are urgently needed. This pilot study aims to identify serum protein biomarkers associated with clinical disease activity in untreated JDM and their response to medical therapy. METHODS: SomaScan® technology screened JDM patients for 1305 proteins at three points: 1) before start of treatment, 2) while on therapy, and 3) after treatment tapering when patients were clinically inactive. To define disease associated biomarkers, SomaScan® data from untreated JDM patients (n = 8) were compared to SomaScan® data from an independent age-matched healthy control group (n = 12). Longitudinal analysis defined treatment responsive proteins at three time points: untreated (7 samples), treated (7 samples), and clinically inactive (6 samples). To confirm the SomaScan® data, a subset of nine candidate proteins (CXCL11, IL-17B, IL-17D, IL-22, CXCL10, MCP-1, ANGPT2, MIF, IL-23) were tested by ELISA after adding 2 JDM (one untreated, one clinically inactive) serum samples to the same group of JDM girls (8 untreated, 7 treated; 7 clinically inactive) as well as with 17 age, gender, matched healthy controls. RESULTS: Comparison of untreated JDM versus healthy controls identified 202 elevated and 49 decreased serum proteins in JDM patients with an adjusted p-value < 0.001. Only 82 out of 251 identified biomarker candidates responded to treatment while 12 out of these 82 proteins returned to their original untreated disease levels upon therapy tapering. The ELISA testing of the untreated samples for nine candidate proteins confirmed previously known biomarkers (CXCL10 or IP-10, CXCL11 or I-TAC and MCP-1) and identified novel biomarkers including IL-22, Angiopoetin-2, and IL-17B in a cross-sectional analysis comparing 8 untreated JDM and 17 age/gender matched controls. The subsequent longitudinal data by ELISA were not concordant for some biomarkers (IL-22 and IL-17B), but the other biomarkers either normalized or rebounded concordantly. CONCLUSIONS: Blood accessible protein biomarkers reflecting JDM pathophysiology were identified; some of them rebounded after therapy was tapered. Further studies bridging these biomarkers to specific clinical features of JDM are required to confirm the clinical utility of these serum protein biomarkers.

Tandem Mass Tag-Based Serum Proteome Profiling for Biomarker Discovery in Young Duchenne Muscular Dystrophy Boys.

Blood-accessible molecular biomarkers are becoming highly attractive tools to assess disease progression and response to therapies in Duchenne muscular dystrophy (DMD) especially in very young patients for whom other outcome measures remain subjective and challenging. In this study, we have standardized a highly specific and reproducible multiplexing mass spectrometry method using the tandem mass tag (TMT) strategy in combination with depletion of abundant proteins from serum and high-pH reversed-phase peptide fractionation. Differential proteome profiling of 4 year-old DMD boys (n = 9) and age-matched healthy controls (n = 9) identified 38 elevated and 50 decreased serum proteins (adjusted P < 0.05, FDR <0.05) in the DMD group relative to the healthy control group. As expected, we confirmed previously reported biomarkers but also identified novel biomarkers. These included novel muscle injury-associated biomarkers such as telethonin, smoothelin-like protein 1, cofilin-1, and plectin, additional muscle-specific enzymes such as UTP-glucose-1-phosphate uridylyltransferase, aspartate aminotransferase, pyruvate kinase PKM, lactotransferrin, tissue alpha-l-fucosidase, pantetheinase, and ficolin-1, and some pro-inflammatory and cell adhesion-associated biomarkers such as leukosialin, macrophage receptor MARCO, vitronectin, galectin-3-binding protein, and ProSAAS. The workflow including serum depletion, sample processing, and mass spectrometry analysis was found to be reproducible and stable over time with CV < 20%. Furthermore, the method was found to be superior in terms of specificity compared to other multiplexing affinity-based methods. These findings demonstrate the specificity and reliability of TMT-based mass spectrometry methods in detection and identification of serum biomarkers in presymptomatic young DMD patients.

Comparison of Serum Pharmacodynamic Biomarkers in Prednisone-Versus Deflazacort-Treated Duchenne Muscular Dystrophy Boys.

Prednisone (Pred) and Deflazacort (Dfz) are commonly used glucocorticoids (GCs) for Duchenne muscular dystrophy (DMD) treatment and management. While GCs are known to delay the loss of ambulation and motor abilities, chronic use can result in onerous side effects, e.g., weight gain, growth stunting, loss of bone density, etc. Here, we use the CINRG Duchenne natural history study to gain insight into comparative safety of Pred versus Dfz treatment through GC-responsive pharmacodynamic (PD) biomarkers. Longitudinal trajectories of SOMAscan® protein data obtained on serum of DMD boys aged 4 to 10 (Pred: n = 7; Dfz: n = 8) were analyzed after accounting for age and time on treatment. Out of the pre-specified biomarkers, seventeen candidate proteins were differentially altered between the two drugs (p < 0.05). These include IGFBP-2 and AGER associated with diabetes complications, and MMP-3 associated with extracellular remodeling. As a follow-up, IGFBP-2, MMP-3, and IGF-I were quantified with an ELISA using a larger sample size of DMD biosamples (Dfz: n = 17, Pred: n = 12; up to 76 sera samples) over a longer treatment duration. MMP-3 and IGFBP-2 validated the SOMAscan® signal, however, IGF-I did not. This study identified GC-responsive biomarkers, some associated with safety, that highlight differential PD response between Dfz and Pred.

Disease-specific and glucocorticoid-responsive serum biomarkers for Duchenne Muscular Dystrophy.

Extensive biomarker discoveries for DMD have occurred in the past 7 years, and a vast array of these biomarkers were confirmed in independent cohorts and across different laboratories. In these previous studies, glucocorticoids and age were two major confounding variables. In this new study, using SomaScan technology and focusing on a subset of young DMD patients who were not yet treated with glucocorticoids, we identified 108 elevated and 70 decreased proteins in DMD relative to age matched healthy controls (p value < 0.05 after adjusting for multiple testing). The majority of the elevated proteins were muscle centric followed by cell adhesion, extracellular matrix proteins and a few pro-inflammatory proteins. The majority of decreased proteins were of cell adhesion, however, some had to do with cell differentiation and growth factors. Subsequent treatment of this group of DMD patients with glucocorticoids affected two major groups of pharmacodynamic biomarkers. The first group consisted of 80 serum proteins that were not associated with DMD and either decreased or increased following treatment with glucocorticoids, and therefore were reflective of a broader effect of glucocorticoids. The second group consisted of 17 serum proteins that were associated with DMD and these tended to normalize under treatment, thus reflecting physiologic effects of glucocorticoid treatment in DMD. In summary, we have identified a variety of circulating protein biomarkers that reflect the complex nature of DMD pathogenesis and response to glucocorticoids.

Serum biomarkers of glucocorticoid response and safety in anti-neutrophil cytoplasmic antibody-associated vasculitis and juvenile dermatomyositis.

Glucocorticoids are standard of care for many chronic inflammatory conditions, including juvenile dermatomyositis (JDM) and anti-neutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV). We sought to define pharmacodynamic biomarkers of therapeutic efficacy and safety concerns of glucocorticoid treatment for these two disorders. Previous proteomic profiling of patients with Duchenne muscular dystrophy (DMD) and inflammatory bowel disease (IBD) treated with glucocorticoids identified candidate biomarkers for efficacy and safety concerns of glucocorticoids. Serial serum samples from patients with AAV (n = 30) and JDM (n = 12) were obtained during active disease, and after treatment with glucocorticoids. For AAV, 8 of 11 biomarkers of the anti-inflammatory response to glucocorticoids were validated (P-value ≤0.05; CD23, macrophage-derived cytokine, interleukin-22 binding protein, matrix metalloproteinase-12, T lymphocyte surface antigen Ly9, fibrinogen gamma chain, angiopoietin-2 [all decreased], and protein C [increased]), as were 5 of 7 safety biomarkers (P-value ≤0.05; afamin, matrix metalloproteinase-3, insulin growth factor binding protein-5, angiotensinogen, leptin [all increased]). For JDM, 10 of 11 efficacy biomarkers were validated (P-value ≤0.05; all proteins except fibrinogen gamma chain) and 6 of 7 safety biomarkers (P-value ≤0.05; AAV proteins plus growth hormone binding protein). The identified efficacy biomarkers may be useful as objective outcome measures for early phase proof-of-concept studies when assessing novel anti-inflammatory drugs in JDM and AAV, and likely in other inflammatory disorders. Similarly, safety biomarkers may also be helpful assessing toxicity of alternatives to glucocorticoids.