Table 2.
Modeling factor options frequently adopted by MAQC-II data analysis teams
| Modeling factor | Option | Original analysis (training => validation)
|
||
|---|---|---|---|---|
| Number of teams | Number of endpoints | Number of models | ||
| Summary and normalization | Loess | 12 | 3 | 2,563 |
| RMA | 3 | 7 | 46 | |
| MAS5 | 11 | 7 | 4,947 | |
| Batch-effect removal | None | 10 | 11 | 2,281 |
| Mean shift | 3 | 11 | 7,279 | |
| Feature selection | SAM | 4 | 11 | 3,771 |
| FC+P | 8 | 11 | 4,711 | |
| T-Test | 5 | 11 | 400 | |
| RFE | 2 | 11 | 647 | |
| Number of features | 0~9 | 10 | 11 | 393 |
| 10~99 | 13 | 11 | 4,445 | |
| ≥1,000 | 3 | 11 | 474 | |
| 100~999 | 10 | 11 | 4,298 | |
| Classification algorithm | DA | 4 | 11 | 103 |
| Tree | 5 | 11 | 358 | |
| NB | 4 | 11 | 924 | |
| KNN | 8 | 11 | 6,904 | |
| SVM | 9 | 11 | 986 | |
Analytic options used by two or more of the 14 teams that submitted models for all endpoints in both the original and swap experiments. RMA, robust multichip analysis; SAM, significance analysis of microarrays; FC, fold change; RFE, recursive feature elimination; DA, discriminant analysis; Tree, decision tree; NB, naive Bayes; KNN, K-nearest neighbors; SVM, support vector machine.