Antibody selection for the IHC algorithm. (A) Correlation between mRNA and IHC expression of GATA3, CCR4, TBX21, and CXCR3 in the training cohort. Positive correlation was observed between mRNA and the corresponding protein expression by IHC. The percentage of positivity was estimated in discrete increments of 10% and correlated with the mRNA gene expression obtained using HG U133 Plus 2 arrays. (B) Distribution of immunostain positivity in the corresponding GEP-defined molecular subtypes. Top panels, The PTCL-GATA3 subtype showed higher positivity for GATA3 and CCR4 immunostains compared with the PTCL-TBX21 subtype, with 50% being the best cutoff to separate the 2 subtypes. Bottom panels, The PTCL-TBX21 subtype showed higher positivity for TBX21 and CXCR3 compared with the PTCL-GATA3 subtype, with most PTCL-TBX21 cases showing ≥20% positivity for TBX21 and CXCR3, and most PTCL-GATA3 cases showing <20% positivity. These cutoffs were subsequently used for the IHC algorithm. (C) Heatmap representation of the GATA3, CCR4, TBX21, and CXCR3 immunostains in GEP-defined subtypes and IHC algorithm classification in training cohort cases. The frequency of the individual immunostain positivity in the training cohort in the PTCL subtypes is shown on the right.