Skip to main content

Table 2 Comparison of screening and clustering results (high noise)

From: A method to identify differential expression profiles of time-course gene data with Fourier transformation

    Without screening With screening
AR(1) parameter Method J Error Sil ARI Error Sil ARI Sensitivity Specificity FDR FNR
p = 0.1 FC* 2 .326 .259 .200 .235 .292 .149 .589 .708 .411 .291
3 .321 .192 .199 .298 .214 .151 .561 .716 .439 .283
4 .321 .151 .194 .320 .166 .156 .553 .722 .447 .278
5 .323 .125 .185 .269 .142 .156 .571 .714 .428 .285
8 .324 .084 .175 .324 .094 .148 .552 .722 .448 .277
GPR**       .483 .779 .221 .517
p = 0.2 FC 2 .343 .253 .164 .234 .291 .117 .567 .697 .432 .302
3 .339 .185 .155 .287 .208 .117 .545 .703 .454 .296
4 .338 .149 .151 .306 .160 .120 .539 .708 .461 .291
5 .337 .125 .146 .261 .138 .120 .555 .702 .445 .297
8 .338 .086 .132 .307 .094 .113 .538 .706 .462 .293
GPR       .536 .677 .323 .463
p = 0.3 FC 2 .359 .248 .128 .284 .290 .090 .546 .681 .453 .318
3 .351 .185 .119 .329 .208 .089 .531 .683 .468 .316
4 .350 .148 .115 .347 .159 .092 .526 .685 .473 .314
5 .350 .127 .108 .304 .137 .088 .537 .681 .462 .318
8 .357 .083 .089 .351 .091 .079 .526 .685 .474 .314
GPR       .584 .572 .427 .415
p = 0.5 FC 2 .383 .246 .073 .330 .284 .053 .517 .632 .482 .367
3 .375 .183 .066 .356 .198 .051 .512 .634 .488 .365
4 .369 .151 .062 .365 .158 .052 .510 .633 .490 .366
5 .369 .126 .056 .338 .137 .046 .514 .633 .485 .367
8 .370 .086 .046 .370 .092 .042 .509 .634 .490 .365
GPR       .646 .409 .590 .353
p = 0.7 FC 2 .395 .248 .035 .356 .275 .030 .505 .504 .495 .422
3 .384 .186 .034 .368 .193 .028 .503 .503 .496 .424
4 .383 .148 .031 .373 .155 .026 .502 .502 .497 .421
5 .381 .125 .028 .358 .134 .024 .504 .504 .496 .419
8 .377 .092 .027 .370 .097 .023 .503 .502 .497 .415
  GPR       .679 .337 .662 .320
  1. * FC: proposed method with Fourier coefficients, **GPR: Gaussian process regression
  2. Comparison of estimation error rate (E), Silhouette width (S) and Adjusted Rand Index (ARI) values of model-based clustering without screening vs with screening with J Fourier coefficients including sensitivity, specificity, FDR and FNR with m = 20 time points. These summaries are based on 500 repetitions of each consisting of 800 curves with AR(1) parameter ρ s with the noise standard deviation σ = 1.5.