Table 2. Genomic signatures that predict the magnitude of the CD8⁺ T cell responses using the DAMIP model.

			DAMIP model predictive signatures
			Train on trial 1, test on trial 2								Train on trial 2, test on trial 1
Gene name	Gene symbol	Gene ID	1	2	3	4	5	6	7	8	1	2	3	4	5	6	7	8	9	10	11	12	13	14
Solute carrier family 2 (facilitated glucose transporter), member 6	SLC2A6	Hs.244378 Day 7	×		×	×	×	×	×	×	×	×	×		×		×			×	×	×	×
Eukaryotic translation initiation factor 2 alpha kinase 4	EIF2AK4	Hs.412102 Day 7	×	×	×		×		×	×								×		×				×
Integrin, alpha L (antigen CD11A)	ITGAL/LFA-1	Hs.174103 Day 7			×									×		×	×	×	×
C-terminal binding protein 1	CTBP1	Hs.208597 Day 7				×										×
Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein	YWHAE	Hs.513851 Day 3				×	×												×			×
Transcribed locus		Hs.619443 Day 7						×			×	×	×	×	×	×	×	×	×	×	×	×	×	×
Protein phosphatase 1, regulatory (inhibitor) PPP1R14A subunit 14A		Hs.631569 Day 3							×															×
Family with sequence similarity 62 member B	FAM62B	Hs.649908 Day 7		×						×				×									×
Transcribed locus		Hs.42650 Day 7									×	×									×
Accuracy of 10-fold cross-validation (%)			93	93	93	93	93	93	93	93	90	90	100	100	100	100	90	90	90	90	90	90	100	100
Accuracy of 1-fold blind prediction (%)			80	80	80	80	80	90	90	90	87	87	80	73	73	73	73	73	73	73	87	73	80	73
Accuracy of 10-fold blind prediction (%)			81	80	81	80	81	85	85	88	84	84	76	72	75	71	73	71	71	75	84	73	76	70

This table summarizes the classification rules that have tenfold cross-validation prediction of at least 80%. Tenfold cross validation on trial 1 resulted in eight different DAMIP predictive signatures, each of which had a tenfold unbiased estimate of 93% prediction rate in trial 1. SLC2A6 and EIF2AK4 are represented in several signatures. Blind prediction of rules developed from trial 1 on trial 2 data produced prediction accuracies in the 80-90% range. Tenfold blind predictions were also carried out to evaluate the consistency of the classification rules obtained by subsets of training data only. Here, for trial 1, rule 1 in the tenfold blind test, nine of the ten resulting rules resulted in 80% correct prediction on the blind data from trial 2, and one rule resulted in 90% correct prediction. Thus, the average unbiased prediction rate on the blind data was 81%. However, when we generated the classification rules using the entire training set, they predicted the blind data with an accuracy of 80% singlefold blind test). Conversely, 14 different discriminatory predictive signatures were obtained when trial 2 was used as the training set, with unbiased classification rates in the range 90–100%. EIF2AK4 and SLC2A6 were also represented in these models. Blind prediction on independent trial 1 yielded 73–87% prediction accuracy. Although gene expression data across various time points were all input into the predictive model, most of the discriminatory signature sets (16 out of 22) consisted of only the day 7 expression relative to the day 0.