Constructing patient features

To better describe patients, we use mutated genes to construct patient features. We find that the frequency of genes mutations is different in different cancers. As illustrated in Fig. 1b, some genes are mutated in all cancers, while others are mutated only in certain cancers. Therefore, we define a weight to balance the effects of different genetic mutations. The weight is defined as:

where u(n)is the total number of gene nmutations in the 14 cancers, and vi(n) is the total number of gene nmutations in cancer vi. We create a new 128-dimensional feature by fusing mutant genes from the same sample in the same cluster. Finally, the identical samples are spliced across all clusters to create a 1,280-dimensional vector that represents patient features.