Re-evaluation of publicly available gene-expression databases using machine-learning yields a maximum prognostic power in breast cancer.

Gene expression signatures refer to patterns of gene activities and are used to classify different types of cancer, determine prognosis, and guide treatment decisions. Advancements in high-throughput technology and machine learning have led to improvements to predict a patient's prognosis for different cancer phenotypes. However, computational methods for analyzing signatures have not been used to evaluate their prognostic power. Contention remains on the utility of gene expression signatures for prognosis. The prevalent approaches include random signatures, expert knowledge, and machine learning to construct an improved signature. We unify these approaches to evaluate their prognostic power. Re-evaluation of publicly available gene-expression data from 8 databases with 9 machine-learning models revealed previously unreported results. Gene-expression signatures are confirmed to be useful in predicting a patient's prognosis. Convergent evidence from [Formula: see text] 10,000 signatures implicates a maximum prognostic power. By calculating the concordance index, which measures how well patients with different prognoses can be discriminated, we show that a signature can correctly discriminate patients' prognoses no more than 80% of the time. Additionally, we show that more than 50% of the potentially available information is still missing at this value. We surmise that an accurate prognosis must incorporate molecular, clinical, histological, and other complementary factors.

Comparative analysis of molecular signatures reveals a hybrid approach in breast cancer: Combining the Nottingham Prognostic Index with gene expressions into a hybrid signature.

The diagnosis of breast cancer-including determination of prognosis and prediction-has been traditionally based on clinical and pathological characteristics such as tumor size, nodal status, and tumor grade. The decision-making process has been expanded by the recent introduction of molecular signatures. These signatures, however, have not reached the highest levels of evidence thus far. Yet they have been brought to clinical practice based on statistical significance in prospective as well as retrospective studies. Intriguingly, it has also been reported that most random sets of genes are significantly associated with disease outcome. These facts raise two highly relevant questions: What information gain do these signatures procure? How can one find a signature that is substantially better than a random set of genes? Our study addresses these questions. To address the latter question, we present a hybrid signature that joins the traditional approach with the molecular one by combining the Nottingham Prognostic Index with gene expressions in a data-driven fashion. To address the issue of information gain, we perform careful statistical analysis and comparison of the hybrid signature, gene expression lists of two commercially available tests as well as signatures selected at random, and introduce the Signature Skill Score-a simple measure to assess improvement on random signatures. Despite being based on in silico data, our research is designed to be useful for the decision-making process of oncologists and strongly supports association of random signatures with outcome. Although our study shows that none of these signatures can be considered as the main candidate for providing prognostic information, it also demonstrates that both the hybrid signature and the gene expression list of the OncotypeDx signature identify patients who may not require adjuvant chemotherapy. More importantly, we show that combining signatures substantially improves the identification of patients who do not need adjuvant chemotherapy.

The Mechanical Fingerprint of Circulating Tumor Cells (CTCs) in Breast Cancer Patients.

Circulating tumor cells (CTCs) are a potential predictive surrogate marker for disease monitoring. Due to the sparse knowledge about their phenotype and its changes during cancer progression and treatment response, CTC isolation remains challenging. Here we focused on the mechanical characterization of circulating non-hematopoietic cells from breast cancer patients to evaluate its utility for CTC detection. For proof of premise, we used healthy peripheral blood mononuclear cells (PBMCs), human MDA-MB 231 breast cancer cells and human HL-60 leukemia cells to create a CTC model system. For translational experiments CD45 negative cells-possible CTCs-were isolated from blood samples of patients with mamma carcinoma. Cells were mechanically characterized in the optical stretcher (OS). Active and passive cell mechanical data were related with physiological descriptors by a random forest (RF) classifier to identify cell type specific properties. Cancer cells were well distinguishable from PBMC in cell line tests. Analysis of clinical samples revealed that in PBMC the elliptic deformation was significantly increased compared to non-hematopoietic cells. Interestingly, non-hematopoietic cells showed significantly higher shape restoration. Based on Kelvin-Voigt modeling, the RF algorithm revealed that elliptic deformation and shape restoration were crucial parameters and that the OS discriminated non-hematopoietic cells from PBMC with an accuracy of 0.69, a sensitivity of 0.74, and specificity of 0.63. The CD45 negative cell population in the blood of breast cancer patients is mechanically distinguishable from healthy PBMC. Together with cell morphology, the mechanical fingerprint might be an appropriate tool for marker-free CTC detection.