Iterative feature removal yields highly discriminative pathways.

BACKGROUND: We introduce Iterative Feature Removal (IFR) as an unbiased approach for selecting features with diagnostic capacity from large data sets. The algorithm is based on recently developed tools in machine learning that are driven by sparse feature selection goals. When applied to genomic data, our method is designed to identify genes that can provide deeper insight into complex interactions while remaining directly connected to diagnostic utility. We contrast this approach with the search for a minimal best set of discriminative genes, which can provide only an incomplete picture of the biological complexity. RESULTS: Microarray data sets typically contain far more features (genes) than samples. For this type of data, we demonstrate that there are many equivalently-predictive subsets of genes. We iteratively train a classifier using features identified via a sparse support vector machine. At each iteration, we remove all the features that were previously selected. We found that we could iterate many times before a sustained drop in accuracy occurs, with each iteration removing approximately 30 genes from consideration. The classification accuracy on test data remains essentially flat even as hundreds of top-genes are removed.Our method identifies sets of genes that are highly predictive, even when comprised of genes that individually are not. Through automated and manual analysis of the selected genes, we demonstrate that the selected features expose relevant pathways that other approaches would have missed. CONCLUSIONS: Our results challenge the paradigm of using feature selection techniques to design parsimonious classifiers from microarray and similar high-dimensional, small-sample-size data sets. The fact that there are many subsets of genes that work equally well to classify the data provides a strong counter-result to the notion that there is a small number of "top genes" that should be used to build classifiers. In our results, the best classifiers were formed using genes with limited univariate power, thus illustrating that deeper mining of features using multivariate techniques is important.

Evaluation of BacLite Rapid MRSA, a rapid culture based screening test for the detection of ciprofloxacin and methicillin resistant S. aureus (MRSA) from screening swabs.

BACKGROUND: Methicillin-resistant Staphylococcus aureus (MRSA) is a major nosocomial pathogen worldwide. The need for accurate and rapid screening methods to detect MRSA carriers has been clearly established. The performance of a novel assay, BacLite Rapid MRSA (Acolyte Biomedica, UK) for the rapid detection (5 h) and identification of hospital associated ciprofloxacin resistant strains of MRSA directly from nasal swab specimens was compared to that obtained by culture on Mannitol salt agar containing Oxacillin (MSAO) after 48 h incubation. RESULTS: A total of 1382 nasal screening swabs were tested by multiple operators. The BacLite Rapid MRSA test detected 142 out of the 157 confirmed MRSA that were detected on MSAO giving a diagnostic sensitivity of 90.4, diagnostic specificity of 95.7% and a negative predictive value of 98.7%. Of the 15 false negatives obtained by the BacLite Rapid MRSA test, seven grew small amounts (< 10 colonies of MRSA) on the MSAO culture plate and five isolates were ciprofloxacin sensitive. However there were 13 confirmed BacLite MRSA positive samples, which were negative by the direct culture method, probably due to overgrowth on the MSAO plate. There were 53 false positive results obtained by the BacLite Rapid MRSA test at 5 h and 115 cases where MRSA colonies were tentatively identified on the MSAO plate when read at 48 h, and which subsequently proved not to be MRSA. CONCLUSION: The BacLite MRSA test is easy to use and provides a similar level of sensitivity to conventional culture for the detection of nasal carriage of MRSA with the advantage that the results are obtained much more rapidly.

Fourier-ring descriptor to characterize rare circulating cells from images generated using immunofluorescence microscopy.

We address the problem of subclassification of rare circulating cells using data driven feature selection from images of candidate circulating tumor cells from patients diagnosed with breast, prostate, or lung cancer. We determine a set of low level features which can differentiate among candidate cell types. We have implemented an image representation based on concentric Fourier rings (FRDs) which allow us to exploit size variations and morphological differences among cells while being rotationally invariant. We discuss potential clinical use in the context of treatment monitoring for cancer patients with metastatic disease.

Which biomarkers reveal neonatal sepsis?

We address the identification of optimal biomarkers for the rapid diagnosis of neonatal sepsis. We employ both canonical correlation analysis (CCA) and sparse support vector machine (SSVM) classifiers to select the best subset of biomarkers from a large hematological data set collected from infants with suspected sepsis from Yale-New Haven Hospital's Neonatal Intensive Care Unit (NICU). CCA is used to select sets of biomarkers of increasing size that are most highly correlated with infection. The effectiveness of these biomarkers is then validated by constructing a sparse support vector machine diagnostic classifier. We find that the following set of five biomarkers capture the essential diagnostic information (in order of importance): Bands, Platelets, neutrophil CD64, White Blood Cells, and Segs. Further, the diagnostic performance of the optimal set of biomarkers is significantly higher than that of isolated individual biomarkers. These results suggest an enhanced sepsis scoring system for neonatal sepsis that includes these five biomarkers. We demonstrate the robustness of our analysis by comparing CCA with the Forward Selection method and SSVM with LASSO Logistic Regression.