dorsal/arxiv
View SchemaA windowed local fdr estimator providing higher resolution and robust thresholds
| Authors | Rishi L. Khan, Rajanikanth Vadigepalli, Guang Gao, James S. Schwaber |
|---|---|
| Categories | |
| ArXiv ID | q-bio/0702044 |
| URL | https://arxiv.org/abs/q-bio/0702044 |
Abstract
Motivation: In microarray analysis, special consideration must be given to the issues of multiple statistical tests and typically p-values are adjusted to control family-wise error rate (FWER) or false discovery rate (FDR). FDR metrics have been suggested for controlling false positives, however, genes with p-values close to the threshold typically have a higher chance of being false positives than genes with very low p-values. The local FDR (fdr) metric gives the number of false positives in the vicinity of a test statistic. We propose a new fdr estimator that uses windows instead of binsand define heuristics that use the fluctuations in the estimator to determine robust thresholds for classifying differential expression. Results: Our fdr approach estimates the false discovery rate within a window of p-values. We present heuristics that derive robust fdr thresholds such that a significant change in the fdr threshold yields a small change in the number of rejected hypotheses. We compare these thresholds with thresholds from other approaches using two simulated datasets and one cancer microarray dataset. In the latter, our estimator finds two robust thresholds. Since our fdr estimator is an extension of the FDR metric, it can be used with many FDR estimation methods. Availability: An R function implementing the proposed estimator is available at http://www.dbi.tju.edu/dbi/tools/fdr Contact: james.schwaber@jefferson.edu Supplementary Information: Supplementary figures and code are available at http://www.dbi.tju.edu/dbi/tools/fdr
{
"annotation_id": "7d552746-8463-4734-a8a7-a78d6f04c4dc",
"date_created": "2026-03-02T18:01:35.566000Z",
"date_modified": "2026-03-02T18:01:35.566000Z",
"file_hash": "430fac00c1d8c6b4c07290cc5053382921af223f507c2453c9b31f7c38a8350f",
"private": false,
"record": {
"abstract": "Motivation: In microarray analysis, special consideration must be given to\nthe issues of multiple statistical tests and typically p-values are adjusted to\ncontrol family-wise error rate (FWER) or false discovery rate (FDR). FDR\nmetrics have been suggested for controlling false positives, however, genes\nwith p-values close to the threshold typically have a higher chance of being\nfalse positives than genes with very low p-values. The local FDR (fdr) metric\ngives the number of false positives in the vicinity of a test statistic. We\npropose a new fdr estimator that uses windows instead of binsand define\nheuristics that use the fluctuations in the estimator to determine robust\nthresholds for classifying differential expression.\n Results: Our fdr approach estimates the false discovery rate within a window\nof p-values. We present heuristics that derive robust fdr thresholds such that\na significant change in the fdr threshold yields a small change in the number\nof rejected hypotheses. We compare these thresholds with thresholds from other\napproaches using two simulated datasets and one cancer microarray dataset. In\nthe latter, our estimator finds two robust thresholds. Since our fdr estimator\nis an extension of the FDR metric, it can be used with many FDR estimation\nmethods.\n Availability: An R function implementing the proposed estimator is available\nat http://www.dbi.tju.edu/dbi/tools/fdr\n Contact: james.schwaber@jefferson.edu\n Supplementary Information: Supplementary figures and code are available at\nhttp://www.dbi.tju.edu/dbi/tools/fdr",
"arxiv_id": "q-bio/0702044",
"authors": [
"Rishi L. Khan",
"Rajanikanth Vadigepalli",
"Guang Gao",
"James S. Schwaber"
],
"categories": [
"q-bio.QM",
"q-bio.GN"
],
"title": "A windowed local fdr estimator providing higher resolution and robust thresholds",
"url": "https://arxiv.org/abs/q-bio/0702044"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "38a050e7-85f8-4fe6-9645-1e3b8460f0f9",
"id": "arXiv Dataset IDs",
"type": "Model",
"variant": "snapshot-2026-03-01",
"version": "0.1.0"
},
"user_id": 1000002
}