Robustness of the accuracy of FAIME-derived Oncogenic FAIME Features of HNSCC in separating tumor from control tissue samples.

57 deregulated Oncogenic FAIME Features of HNSCC were reproducibly derived by FAIME in the three independent datasets (A, B, and C; Figure 3, Table 1) and shown that they can jointly classify non-tumor control tissue from HNSCC tumors in two validation datasets (D and E; Figure 5, Table 1). This table also illustrates that the constituents of the HNSCC Oncogenic FAIME Features of HNSCC (columns), the 24 GO Molecular Functions and the 33 KEGG pathways, were also independently accurate in separating tumor from control tissue samples in the validation datasets. Furthermore, we report that this predictive accuracy is robustly reproduced using five different unsupervised clustering algorithms (rows below). In the remainder of the manuscript, these Oncogenic FAIME Features of HNSCC are pooled together for prognosis predictions in Figures 5–6.*Default parameters of the clustering algorithms were used unless otherwise specified. After applying each clustering method to the FPHNSCC,s, the first two-way partition of samples was used to determine two classes compared with the tumor sample group and the control group of two independent datasets D and E (Table 1). False positive results were control samples misclassified within the tumor sample cluster and false negative results were tumor samples classified as control tissue. The FAIME-Score accuracy for separating 23 HNSCC tumors from 13 control samples of the E-MEXP44_hu6800 array was calculated as .§: GO-MF terms and KEGG pathways were not included if there were no mapped genes that passed the IQR filter in this validation dataset.Legend: FPHNSCC,s is defined as the subset of the Functional FAIME Profile consisting of the 57 Oncogenic FAIME Features of HNSCC deregulated in the initial datasets A, B and C (Equation 4, Methods, Figure 3); GO: Gene ontology, MF: molecular function.