TABLE 2.
Features | CON versus NSCLCa | CON versus I_IIIa | I_III versus DMa | CON versus DMa | |
---|---|---|---|---|---|
Sputum | All | 0.750 | 0.850 | 0.767 | 0.744 |
DAMb | 0.728 | 0.867 | 0.742 | 0.756 | |
RFEc | 0.803 | 0.850 | 0.708 | 0.739 | |
Gut | All | 0.760 | 0.817 | 0.760 | 0.650 |
DAMb | 0.515 | —d | — | — | |
RFEc | 0.692 | 0.717 | 0.673 | 0.720 | |
Sputum + gut | All | 0.825 | 0.896 | 0.754 | 0.866 |
RFEc | 0.783 | 0.854 | 0.691 | 0.804 |
In each comparison (e.g., CON versus NSCLC), the modeling strategies with the highest and lowest performance as measured by AUC values are highlighted in bold font and gray shading, respectively.
DAM, differentially abundant microbes. In more than 90% of tests, there were too few or no differentially abundant genera, so it was considered that the DAM model was not suitable for fecal microbiota.
RFE, features with top-ranked contributions to the models were used (see Materials and Methods).
—, indicates no data.