br Although the focus of CHASMplus is

Although the focus of CHASMplus is driver missense muta-tions, of the 240 Ruxolitinib containing at least one predicted driver missense mutation, 75 genes may be novel. They were not pre-viously included in either the Cancer Gene Census (Forbes et al., 2017) or our previous work with the TCGA PancanAtlas (Bailey et al., 2018) (Table S4). Based on gene ontology enrichment (Huang et al., 2008), these genes occurred in pathways related to known hallmarks of cancer (Hanahan and Weinberg, 2011), such as biological processes for DNA damage response, cell cycle, transcriptional regulation, cell-cell adhesion, and cell dif-ferentiation (Table S4; STAR Methods). These results suggest that CHASMplus has the potential to discriminate driver and passenger mutations in both well-known and putative cancer- 
driver genes, although follow-up experiments are required for confirmation.
CHASMplus Identifies Both Common and Rare Cancer-Driver Mutations
Previous driver gene studies have suggested that there are few common driver genes and many rare driver genes (long-tail hy-pothesis [Ding et al., 2010; Garraway and Lander, 2013)]). How-ever, the overall mutation frequency of a gene does not account for the confounding presence of passenger mutations within a driver gene. Based on our mutation-level analysis, we observed that the spectrum of rare (<1% of cancer samples), intermediate (1%–5%), and common (>5%) frequency driver missense muta-tions varied substantially among cancer types (Figure 3A). For example, uveal melanoma (UVM) was dominated by common driver missense mutations (88%), while head and neck squa-mous cell carcinoma (HNSC) was dominated by rare driver missense mutations (63%). However, when we considered all cancer types together (pan-cancer), the overall proportion of rare driver missense mutations considered rare was slightly greater than for common (35.5% and 35.4%, respectively) and 4-fold greater than found by the cancer hotspots method (8%,
p < 2.2e 16, Fisher’s exact test) (Chang et al., 2016). These re-sults suggest that rare driver missense mutations have a greater role in many cancer types than previously suggested but that this might not be the case for every cancer type. We observed, after adjusting for sample size, that the average tumor mutation burden for a cancer type positively correlated with the preva-
While cancer types appear to have different proportions of rare, intermediate, and common driver missense mutations across cancer types, this result could be confounded by differ-ences in subtypes or the cell of origin. A common driver mutation in an uncommon subtype could be perceived, overall, as rare. To test this, we analyzed whether driver missense mutations within a gene showed noticeable enrichment for samples of a particular subtype. In TCGA, there are 12 cancer types with available sub-type information (Sanchez-Vega et al., 2018). Subtype specificity partially explained differences in driver mutation frequency spec-trum between cancer types. Fifty-five out of 223 genes (24.7%) contained putative driver missense mutations that were enriched in particular cancer subtypes (q value % 0.1; chi-square test; Figure 3C; Table S5; Figure S4). Several genes showed strong specificity, consistent with prior literature, such as NFE2L2 mutations in esophageal cancers of squamous cells of origin (Cancer Genome Atlas Research Network et al., 2017), TP53 mutations in human-papillomavirus-negative tumors in head and neck cancer (Cancer Genome Atlas Network, 2015), KIT mutations in testicular seminomas (Kemmer et al., 2004) and PIK3CA mutations in the Luminal A subtype of breast cancer (Cancer Genome Atlas Network, 2012b). It should be noted that in some cases, these differences are confounded by the fact that subtypes were originally defined by mutation status (GBM or LGG with IDH1/IDH2 mutations).