Figure 1.
Graphical Depiction of HLA Restriction Model
Graphical depiction of the model used to infer HLA-restricted epitopes from ELISpot data. The probability of each peptide having a reaction is parameterized by a noisy-OR distribution over all of the HLA alleles it is connected to (Equations 1 and 2). The values of the HLA and peptide nodes are observed for each donor, and we are interested in finding which qij > 0—that is, which arcs are present in the graphical model. Each person has between three and six distinct HLA class 1alleles. Thus, for a given donor, between three and six HLA nodes will be “on” (hi = 1).
Figure 2.
Actual versus Estimated FDR (A) and False Negatives versus False Positives (B)
Results from using our model selection procedure and FDR estimation procedure on three datasets generated from a synthetic model learned on the real HIV data. There is a one-to-one correspondence between the points plotted in each figure.
(A) Estimated and actual FDR. The dashed line denotes the idealized curve.
(B) The number of false negatives (qij not recovered in these experiments, but appearing in the synthetic model), compared with the number of false positives (qij recovered in these experiments, but not in the synthetic model).
Table 1.
Previously Known Promiscuity
Table 2.
Promiscuity Updated with Present Analysis