Fig. 2: Small numbers of the best features are sufficient for accurate cancer classification in TCGA cases: the relationship between numbers of features and diagnostic accuracies in pan-cancer models.
From: Minimalist approaches to cancer tissue-of-origin classification by DNA methylation
Each point represents a single classifier from the (a) X vs. all, (b) pairwise differential, or (c) hybrid series. The smallest models are indicated; the smallest hybrid model is shown in greater detail in Fig. 3. The colored trend lines for the plots are rectangular hyperbolas with asymptotes (maximum predicted accuracies) of 94.9% (blue), 96.8% (red), and 96.9% (green) for the X vs. all, pairwise, and hybrid series, respectively. Overall, little additional accuracy is gained by the incorporation of additional highly-ranked probes.
