Skip to main content

Table 1 Training data size affects fRMA reproducibility

From: Thawing Frozen Robust Multi-array Analysis (fRMA)

   Batch Size     
   3 5 10 15 20
Number of Batches 5 0.7585 (0.1574) 0.7573 (0.1437) 0.8515 (0.1435) 0.5812 (0.1231) 0.4439 (0.0916)
  10 0.6795 (0.0799) 0.7173 (0.0858) 0.6563 (0.0641) 0.4506 (0.1073) 0.3878 (0.0901)
  20 0.5696 (0.0654) 0.4691 (0.0523) 0.5180 (0.0506) 0.4551 (0.0561) 0.3299 (0.0629)
  30 0.4429 (0.0491) 0.3884 (0.0482) 0.3387 (0.0380) 0.3697 (0.0537) 0.3036 (0.0440)
  40 0.3290 (0.0450) 0.3700 (0.0488) 0.2598 (0.0368) 0.2642 (0.0303)  
  50 0.3093 (0.0424) 0.3107 (0.0339) 0.2307 (0.0291)   
  60 0.2661 (0.0374) 0.2454 (0.0322) 0.1955 (0.0261)   
  70 0.2529 (0.0322) 0.2286 (0.0295) 0.2089 (0.0261)   
  80 0.2256 (0.0281) 0.2098 (0.0285) 0.1616 (0.0259)   
  90 0.1922 (0.0274) 0.2058 (0.0248) 0.1566 (0.0163)   
  100 0.1891 (0.0277) 0.1976 (0.0261) 0.1128 (0.0167)   
  1. Median and IQR of the across-replicate median absolute deviations (MAD) for different batch sizes (columns) and number of batches (rows) used to train the fRMA algorithm. The median provides an estimate of the typical MAD; the IQR provides an estimate of the variability seen in MADs across replicates.