Predicting defects in imbalanced data using resampling methods: an empirical investigation.

The development of correct and effective software defect prediction (SDP) models is one of the utmost needs of the software industry. Statistics of many defect-related open-source data sets depict the class imbalance problem in object-oriented projects. Models trained on imbalanced data leads to inaccurate future predictions owing to biased learning and ineffective defect prediction. In addition to this large number of software metrics degrades the model performance. This study aims at (1) identification of useful metrics in the software using correlation feature selection, (2) extensive comparative analysis of 10 resampling methods to generate effective machine learning models for imbalanced data, (3) inclusion of stable performance evaluators-AUC, GMean, and Balance and (4) integration of statistical validation of results. The impact of 10 resampling methods is analyzed on selected features of 12 object-oriented Apache datasets using 15 machine learning techniques. The performances of developed models are analyzed using AUC, GMean, Balance, and sensitivity. Statistical results advocate the use of resampling methods to improve SDP. Random oversampling portrays the best predictive capability of developed defect prediction models. The study provides a guideline for identifying metrics that are influential for SDP. The performances of oversampling methods are superior to undersampling methods.

Clinical Evaluation of a New Point-of-Care System for Chemistry Panel Testing.

BACKGROUND: GEM Premier ChemSTAT™ is a point-of-care (POC) system that measures Na+, K+, Ca++, Cl-, glucose, hematocrit, creatinine, blood urea nitrogen (BUN), tCO2, pH, pCO2, and lactate from a single whole blood specimen, providing rapid results in POC settings such as the emergency department (ED). Accurate measurements of creatinine in whole blood and reporting of estimated glomerular filtration rate (eGFR) can minimize adverse effects of contrast-induced nephropathy. METHODS: Heparinized whole blood specimens from the ED were analyzed on the ChemSTAT by POC staff. Method comparison was performed against the cobas Integra c501 for creatinine, BUN, and tCO2, and against the GEM Premier 4000 for all other analytes. Precision was conducted with whole blood specimens assayed in triplicate over 6 days. Creatinine results from whole blood and plasma were used for eGFR, by isotope dilution mass spectrometry-traceable Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration equations, and eGFR concordance was assessed. RESULTS: Creatinine, BUN, and tCO2 correlated well with plasma samples on the cobas, and all other analytes correlated well with whole blood specimens on the GEM Premier 4000 across the tested sample ranges. The regression slope was 0.951 to 1.047, along with a correlation coefficient (r) of ≥0.982 for all analytes. The pooled within-sample precision was 0% to 2.5% for all analytes. CONCLUSIONS: ChemSTAT demonstrated a strong correlation with the comparative methods and excellent precision. The system's analytical performance and continuous quality management make it suitable for use in the ED to provide rapid reliable test results, which could minimize the time to treatment and improve ED efficiency.

DSG3 as a biomarker for the ultrasensitive detection of occult lymph node metastasis in oral cancer using nanostructured immunoarrays.

OBJECTIVES: The diagnosis of cervical lymph node metastasis in head and neck squamous cell carcinoma (HNSCC) patients constitutes an essential requirement for clinical staging and treatment selection. However, clinical assessment by physical examination and different imaging modalities, as well as by histological examination of routine lymph node cryosections can miss micrometastases, while false positives may lead to unnecessary elective lymph node neck resections. Here, we explored the feasibility of developing a sensitive assay system for desmoglein 3 (DSG3) as a predictive biomarker for lymph node metastasis in HNSCC. MATERIALS AND METHODS: DSG3 expression was determined in multiple general cancer- and HNSCC-tissue microarrays (TMAs), in negative and positive HNSCC metastatic cervical lymph nodes, and in a variety of HNSCC and control cell lines. A nanostructured immunoarray system was developed for the ultrasensitive detection of DSG3 in lymph node tissue lysates. RESULTS: We demonstrate that DSG3 is highly expressed in all HNSCC lesions and their metastatic cervical lymph nodes, but absent in non-invaded lymph nodes. We show that DSG3 can be rapidly detected with high sensitivity using a simple microfluidic immunoarray platform, even in human tissue sections including very few HNSCC invading cells, hence distinguishing between positive and negative lymph nodes. CONCLUSION: We provide a proof of principle supporting that ultrasensitive nanostructured assay systems for DSG3 can be exploited to detect micrometastatic HNSCC lesions in lymph nodes, which can improve the diagnosis and guide in the selection of appropriate therapeutic intervention modalities for HNSCC patients.

Ultrasensitive detection of cancer biomarkers in the clinic by use of a nanostructured microfluidic array.

Multiplexed biomarker protein detection holds unrealized promise for clinical cancer diagnostics due to lack of suitable measurement devices and lack of rigorously validated protein panels. Here we report an ultrasensitive electrochemical microfluidic array optimized to measure a four-protein panel of biomarker proteins, and we validate the protein panel for accurate oral cancer diagnostics. Unprecedented ultralow detection into the 5-50 fg·mL(-1) range was achieved for simultaneous measurement of proteins interleukin 6 (IL-6), IL-8, vascular endothelial growth factor (VEGF), and VEGF-C in diluted serum. The immunoarray achieves high sensitivity in 50 min assays by using off-line protein capture by magnetic beads carrying 400,000 enzyme labels and ~100,000 antibodies. After capture of the proteins and washing to inhibit nonspecific binding, the beads are magnetically separated and injected into the array for selective capture by antibodies on eight nanostructured sensors. Good correlations with enzyme-linked immunosorbent assays (ELISA) for protein determinations in conditioned cancer cell media confirmed the accuracy of this approach. Normalized means of the four protein levels in 78 oral cancer patient serum samples and 49 controls gave clinical sensitivity of 89% and specificity of 98% for oral cancer detection, demonstrating high diagnostic utility. The low-cost, easily fabricated immunoarray provides a rapid serum test for diagnosis and personalized therapy of oral cancer. The device is readily adaptable to clinical diagnostics of other cancers.

Sequential layer analysis of protein immunosensors based on single wall carbon nanotube forests.

Electrochemical immunosensors using vertically aligned single wall carbon nanotube (SWNT) forests can provide ultrasensitive, accurate cancer biomarker protein assays. Herein we report a systematic investigation of the structure, thickness, and functionality of each layer of these immunosensors using atomic force microscopy (AFM), quartz crystal microbalance (QCM), and scanning white light interferometry (SWLI). This provides a detailed picture of the surface morphology of each layer along with surface concentration and thickness of each protein layer. Results reveal that the major reasons for sensitivity gain can be assigned to the dense packing of carboxylated SWNT forest tips, which translate to a large surface concentration of capture antibodies, together with the high quality of conductive SWNT forests.

Electrochemical immunosensors for antibodies to peanut allergen ara h2 using gold nanoparticle-peptide films.

Life-threatening allergies to peanuts and tree nuts can be revealed by detecting antibodies (IgEs) to their allergens in patient serum. Herein, we compare several immunosensor-like methodologies for sensitive detection of antibodies to a peptide sequence from the major peanut allergen, Arachis hypogaea 2 (Ara h2). The sensors feature a synthetic peptide layer of the major IgE-binding epitope from Ara h2 attached to a dense gold nanoparticle (AuNP) film on a pyrolytic graphite (PG) electrode. The AuNP-peptide sensor was used to determine model chicken antipeanut antibodies (IgY) in serum. Faradaic and nonfaradaic impedance strategies were compared to amperometric detection. Measurements employed goat antichicken secondary antibodies (Ab(2)) labeled with horseradish peroxidase (HRP) to bind to IgY on the sensor and provide amplified signals. The best impedimetric sensor configuration featured HPR-catalyzed precipitation of the enzyme product onto the sensor measured by nonfaradaic impedance. This sensor configuration had the best detection limit (DL) of 5 pg mL(-1) and the best linear range of over 5 orders of magnitude (from 5 pg mL(-1) to 1 microg mL(-1)) for IgY antibody in undiluted calf serum. This DL was 100-fold lower than label-free impedimetric immunosensors (0.5 ng mL(-1)) and 60-fold lower than when using HRP-Ab(2) in amperometric immunosensors (0.3 ng mL(-1)).

Ultrasensitive electrochemical immunosensor for oral cancer biomarker IL-6 using carbon nanotube forest electrodes and multilabel amplification.

Squamous cell carcinomas of head and neck (HNSCC) are associated with immune, inflammatory, and angiogenic responses involving interleukin-6 (IL-6). This article reports an ultrasensitive electrochemical immunosensor for human IL-6 and proof-of-concept studies of IL-6 detection in HNSCC cells. Single wall carbon nanotube (SWNT) forests with attached capture antibodies (Ab(1)) for IL-6 were used in an electrochemical sandwich immunoassay protocol using enzyme label horseradish peroxidase (HRP) to measure very low (

Single-wall carbon nanotube forest arrays for immunoelectrochemical measurement of four protein biomarkers for prostate cancer.

Protein arrays that measure multiple protein cancer biomarkers in clinical samples hold great promise for reliable early cancer detection. Herein, we report a prototype 4-unit electrochemical immunoarray based on single-wall carbon nanotube forests for the simultaneous detection of multiple protein biomarkers for prostate cancer. Immunoarray procedures were designed to measure prostate specific antigen (PSA), prostate specific membrane antigen (PSMA), platelet factor-4 (PF-4), and interleukin-6 (IL-6) simultaneously in a single serum sample. All of these proteins are elevated in serum of patients with prostate cancer, but they have widely different relative levels of serum concentration. Horseradish peroxidase (HRP) was used as label on detection (secondary) antibodies in a sandwich immunoassay scheme. Biotinylated secondary antibodies (Ab(2)) that bind specifically to streptavidin-HRP conjugates provided 14-16 labels per antibody and gave the necessary higher sensitivity required for PF-4 and IL-6 detection at physiological levels. Conventional singly labeled Ab(2)-HRP conjugates were sufficient for PSA and PSMA detection. Immunoarrays were used to measure four biomarkers in clinical human serum samples of prostate cancer patients and controls with excellent correlation to referee enzyme-linked immunosorbent (ELISA) assays.

Electrochemical Immunosensors for Interleukin-6. Comparison of Carbon Nanotube Forest and Gold Nanoparticle platforms.

Electrochemical immunosensors based on single wall nanotube (SWNT) forests and 5 nm glutathione-protected gold nanoparticles (GSH-AuNP) were developed and compared for the measurement of human cancer biomarker interleukin-6 (IL-6) in serum. Detection was based on sandwich immunoassays using multiple (14-16) horseradish peroxidase labels conjugated to a secondary antibody. Performance was optimized by effective blocking of non-specific binding (NSB) of the labels using bovine serum albumin. The GSH-AuNP immunosensor gave a detection limit (DL) of 10 pg mL(-1) IL-6 (500 amol mL(-1)) in 10 muL calf serum, which was 3-fold better than 30 pg mL(-1) found for the SWNT forest immunosensor for the same assay protocol. The GSH-AuNPs platform also gave a much larger linear dynamic range (20-4000 pg mL(-1)) than the SWNT system (40-150 pg mL(-1)), but the SWNTs had 2-fold better sensitivity in the low pg mL(-1) range.