TABLE 3.
PLS-based feature selection of cytokines that differentiate groups of children with severe acute malnutrition who had diarrhea or died1
Diarrhea |
Death |
||||
Cytokine | Component 1 (Cor: 0.49; R2: 0.24; Q2: 0.08)2 | Feature stability,3 % | Component 1 (Cor: 0.60; R2: 0.36; Q2: 0.18)2 | Feature stability,3 % | |
1 | EGF | 0.314 | — | 0.15 | — |
2 | Eotaxin | 0.14 | — | −0.03 | — |
3 | GCSF | 0.644 | 100 | 0.684 | 90 |
4 | GMCSF | 0.534 | — | 0.414 | — |
5 | IFNα2 | 0.594 | 100 | 0.484 | 70 |
6 | IFNγ | 0.394 | — | 0.17 | — |
7 | IL10 | 0.26 | — | 0.29 | — |
8 | IL12p40 | 0.26 | — | 0.11 | — |
9 | IL12p70 | 0.644 | 100 | 0.554 | 80 |
10 | IL13 | 0.354 | — | 0.54 | 100 |
11 | IL15 | 0.594 | — | 0.584 | 70 |
12 | IL17A | 0.08 | — | 0.19 | — |
13 | IL1RA | 0.654 | 50 | 0.754 | 100 |
14 | IL1α | 0.16 | — | 0.12 | — |
15 | IL2 | 0.744 | 100 | 0.744 | 100 |
16 | IL5 | 0.24 | — | 0.17 | — |
17 | IL6 | 0.564 | 100 | 0.614 | 100 |
18 | IL7 | 0.26 | — | 0.364 | — |
19 | IL8 | 0.34 | — | 0.44 | — |
20 | IP10 | 0.28 | — | 0.14 | — |
21 | MCP1 | 0.344 | — | 0.334 | — |
22 | MIP1α | 0.21 | — | 0.02 | — |
23 | MIP1β | 0.07 | — | 0.03 | — |
24 | TNF-α | 0.374 | 100 | 0.334 | 100 |
25 | TNF-β | 0.564 | 100 | 0.554 | 100 |
26 | VEGF | 0.24 | — | 0.23 | — |
IL3, IL4, and IL1β were not analyzed because they were undetected in most samples. EGF, epidermal growth factor; GCSF, granulocyte-colony stimulating factor; GMCSF, granulocyte-macrophage colony stimulating factor; IFN, interferon; IP, induced protein; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; PLS, partial least square; VEGF, vascular endothelial growth factor.
Cor indicates the correlation strength between the PLS-component 1 and either diarrhea or death. R2 indicates the variance explained by component 1. Q2 indicates the predictive quality of component 1; Q2 is equal to 1 minus the prediction error sum of squares divided by the total sum of squares of the response variable; negative Q2 values indicate that the component is not predictive.
Feature stability indicates the percentage of times that a cytokine was selected as a top-10 feature by using sparse PLS with 10-fold cross-validation.
Correlation >0.3 with component 1.