The threshold value for patient stratification derived from TCGA was 0. This value was almost identical to that of the breast cancer cohort 0. After stratifying patients into predicted responder and non-responder groups at the threshold value, significant survival differences were found in TCGA and Dressman, but not in UVA PPVs of prediction scores were consistently higher than pCR rates in all ovarian and breast cancer cohorts Supplementary Figure As these targeted drugs have not been used in the other cancer types, no patient data sets were available for our direct validation of CONCORD predictions on these cancer types.
However, we found that the efficacy of the drug has already been confirmed for these cancer types by several ongoing studies. Erlotinib has been approved for the advanced pancreatic cancer patients who have not received previous chemotherapy [ 48 ]. A phase 2 study of erlotinib in patients with metastatic colorectal cancer reported that more than one third of evaluable patients had stable disease with favorable toxicity profiles [ 49 ].
A recent randomized, open-label, phase 3 trial explored erlotinib plus Bevacizumab as a new non-chemotherapy-based maintenance option as a first line treatment for patients with unresectable metastatic colorectal cancer previously exposed to bevacizumab-based induction therapy [ 50 ]. Furthermore, a recent phase 2 study of bladder cancer reported that Erlotinib had beneficial effects in short-term clinical outcomes for patients with invasive bladder cancer [ 51 ].
As for vemurafenib, we derived drug sensitivity biomarkers from NCI cancer cell line panel and used gene expression of 19 melanoma patients bearing the target mutation of Vemurafenib B-RAF VE mutation to calculate bCOXEN between melanoma and other cancer types. Again, no patient set treated with this drug was available for our direct validation with these selected cancer types.
Furthermore, there were anecdotal responses among patients with anaplastic thyroid cancer, ovarian cancer, and colorectal cancer [ 52 ]. The CMap had reference profiles for paclitaxel and doxorubicin out of chemotherapy drugs. Thus, we queried our final sixteen paclitaxel biomarkers as a gene signature, which resulted in paclitaxel as top th out of 6, instances top 1.
In subsequent query of 50 final doxorubicin biomarkers, doxorubicin also ranked high at top 73 rd 1. We showed that the cross-cancer CONCORD prediction of several standard chemotherapy agents were significantly predictive for patient responses in different cancer types in an independent, prospective manner. In particular, when its bCOXEN was similar to wCOXEN in the original cancer type, we were able to directly use the predefined threshold from the original cancer to stratify patient outcomes in the second cancer type.
We found that such biomarkers could retain a concordant predictive power across different cancer types based on their consistent gene co-expression networks across the three cancer systems. Hence, it will be interesting to examine novel opportunities to use these drugs on those cancer types by using CONCORD to predict patients with highest probabilities of responding. There have been studies to effectively infer potential drug indications by either matching drug or disease gene expression profiles.
First, CONCORD is designed to provide not only new target cancer types for drug repositioning, but also an accurate statistical prediction model to select responsive patients with the new target cancer type. The latter distinction will be highly important for introducing a novel drug in clinical settings.
It is worth to note several limitations of our current study. It will therefore be very useful if inferences on single drug contributions can be made for patients who were treated with combination therapies. Yet this is challenging for several reasons: First, different chemotherapy drug effects are often correlated, so that it is difficult to decompose them solely into exclusive individual drug effects.
Also, it is difficult to obtain and validate equally highly predictive biomarker models for all single drugs, and individual drug effects cannot be accurately estimated from their combination signature.
Browse Profiles & Photos of Gay Single Men in Concord, NH! Join chriselsom.com, the leader in online dating with more dates, more relationships and more marriages than any other dating site. Jamey is single in Concord, NH USA Free Online Dating, Gay Singles, Gay Dating, Jewish Men, Lesbian Singles, Single Parents. DateHookup is dating sites no credit to practice. Signup and in Guatemala. Matchopolis is the only free dating site Gay Speeddating East Concord (United States.
Furthermore, single and combination drug effects are associated with many other confounding factors such as target patient population and specific clinical settings of each study. However, if high-performing single drug biomarker models for all drugs used in a specific combination regimen can be curated, this opens the possibility of using multivariate logistic regression models on the single drug signatures. This statistical model can then provide single drug effect coefficients, risk odds ratios, and p-values, which may provide, in part, information necessary for evaluating individual drug effects [ 21 , 38 ].
These strategies are currently being investigated. While our CONCORD attempted to utilize de novo molecular information for our cross-cancer prediction, the latter molecular information such as drug activities and resistance under certain cancer-specific mutations will not be apparent until the drug is actually used.
Thus it will be important to obtain and integrate cancer-specific mutation and other molecular information to more accurately predict cross-cancer patient therapeutic responses. We also found that it was difficult to discover consistent biomarkers from different RNA sources, e. It was important to overcome such technical differences by using appropriate quality control and normalization analysis procedures for our CONCORD prediction.
In our current study, developments and validations were mainly applied to patient sets profiled with several Affymetrix microarray platforms. While some of our multi-gene biomarker models have been successfully applied with considerably different platforms, e. Also, our biomarker discovery and modelling methods are highly dependent on the cancer cell line panel and patient data resources in the initial cancer type; thus, our CONCORD approach is currently restricted to drugs and cancer types for which such rich datasets are available.
However, one may need to carefully examine if large reliable patient data resources are available and whether predictive therapeutic biomarkers can be obtained from such molecular data. Likewise, since cultured cell lines can show quite different expression profiles across many key genes, it may be possible to substitute cell line-based expression data from other sources such as patient-derived xenograft tumors, treated metastatic tumors, or other model systems including ex vivo spheroid and autochthonous models.
The list of cancer cell line panels and cancer patient cohorts treated with corresponding cancer therapeutic agents is summarized in Supplementary Table 6 with its cancer types and roles in our CONCORD development.
Patients in all ovarian cancer cohorts were treated with platinum-based systematic chemotherapy with Taxane [ 35 ]. We have not used them for our drug sensitivity biomarker discovery and prediction modeling in any manner. In this study we used breast cancer as a primary cancer type since multiple large patient sets were available.
Win Hindle: Yes, that's right. Want to write a review? Sarah Hindle: As I said to him one day—or you asked him—and said, "Are we more involved than some people who build houses? When guests stay at the property, they check out how quiet the room is, how friendly the staff is, and more. The Numbers.
These sets included important parameters such as pathologic clinical response after chemotherapy which was required for our independent model evaluation and optimal threshold derivation for cross-cancer patient stratification. Other cohorts of cancer patients treated with the drug of interest were used later for external validation of the final prediction model of drug response.
In vitro drug activity and gene expression data of cancer cell lines were used to screen the most accurate drug sensitivity biomarkers for a given drug. The basic unit of biomarker is an individual probe set in the microarray data of cancer cell lines. The drug sensitivity of each biomarker was then represented and prioritized by estimating either correlation coefficient of its gene expressions with drug sensitivity profiles or by independent two sample t -test statistics comparing gene expression levels between highly-sensitive and -resistant cell lines to the drug while controlling false discovery rate FDR at 0.
These were ranked by strength of drug sensitivity in terms of absolute correlation-test or t -test statistics in a descending order. When activity data of a drug were available on multiple cancer cell lines panels, we chose a cancer cell line panel which yielded the largest number of significant gene probe sets in correlation and t -test analyses. We performed our statistical data analysis using the open-source statistical software R v3. Using expression data within each of two systems separately, we constructed two correlation matrices of dimension n x n for n chemosensitivity biomarkers.
The two correlation matrices, e. Then, rc j is derived as. Therefore, rc j reflects the degree of concordance between the NCI and BR panels for expression relationships of probe j with other n-1 probes. If rc j exceeded a cut-off criterion e. Note that since probe j was initially selected among n chemosensitivity biomarkers, it also retained drug sensitivity characteristics.
The former was calculated for any pairs of patient cohorts of a pre-specified original cancer type. The latter was calculated between the original cancer type cohort and a different cancer type cohort. To calculate bCOXEN, we defined the COXEN set for each cancer type as the largest patient cohort available with gene expression microarray data obtained before any cancer-related therapy.
To examine this criterion, we compared these bCOXENs to wCOXEN with the lowest median among gene expression data sets of the original cancer type by using one-sided Wilcoxon rank sum test at a significance level of 0. Once a different cancer type was selected for CONCORD modeling, we further triaged the initial in vitro predictive biomarkers of drug response into the ones that retained highly significant COXEN coefficients across all the three pairs among cancer cell line, original cancer, and different cancer patient sets defined as COXEN sets in Supplementary Table 6. This step was intended to screen drug sensitivity biomarkers that were concordantly co-expressed across all three cancer systems—cell lines, original cancer site, and second cancer site.
We compared and intersected three sets of biomarkers with significant COXEN coefficients in each of the three pairs to identify significantly co-expressed common biomarkers across NCI, breast cancer, and ovarian cancer. This process can also be considered as a humanization step of the initial in vitro drug sensitivity biomarkers.
Therefore, this was one of the key steps that enabled us to obtain the biomarkers to convey drug response information from one cancer type to another cancer type. We performed a principal component regression analysis using a statistical dimension reduction technique to avoid model overfitting due to a large number of biomarkers in the models. Competing models with different numbers of biomarkers were fitted from the most significant CONCORD biomarkers in stratifying in vitro drug sensitivity of cancer cell lines.
To select the optimal prediction model for each drug, we independently evaluated the performance of competing models on the original cancer cohort with the largest number of patients treated with the drug.
The best prediction model was then selected with most stable and significant prediction performance among competing models on the independent patient set of the original cancer type. For future response stratification i.
The final prediction model with the identical threshold value was used to stratify patients of all independent cohorts of both original and different cancer types, prospectively. PPV was then compared with the observed pathological complete response rate pCR under the current standard of care for each cohort in which patients were unselectively treated.
Statistical significance was obtained by a binomial proportion test for the difference between the expected PPV and observed pCR rate. For patient cohorts with survival outcome data, survival distributions were also compared between predicted positive and negative patient groups by Kaplan-Meier survival analysis with a log-rank test. This work was supported by Moffitt Research Fund of H. Lee, the Biostatistics Core Facility at the H. Read article at publisher's site DOI : BMC Bioinformatics , 21 1 , 06 Jul Cited by: 0 articles PMID: Transl Lung Cancer Res , 8 2 , 01 Apr Free to read.
This data has been text mined from the article, or deposited into data resources.