Combining Algorithms to Find Signatures That Predict Risk in Early-Stage Stomach Cancer

This study applied two mathematical algorithms, lattice up-stream targeting (LUST) and D -basis, to the identification of prognostic signatures from cancer gene expression data. The LUST algorithm looks for metagenes, which are sets of genes that are either overexpressed or underexpressed in the same patients. Whereas LUST runs unsupervised by clinical data, the D -basis algorithm uses implications and association rules to relate gene expression to clinical outcomes. The D -basis selects a small subset of the metagene (a signature) to predict survival. The two algorithms, LUST and The first signature (DU4) consists of genes that are underexpressed on the long-survival/low-risk group: The metagenes associate with TCGA cluster C1. Both our signatures and cluster C1 identify tumors that are genomically silent, and have a low mutation load or mutation count. Furthermore, our signatures identify tumors that are predominantly in the WHO classification of poorly cohesive and the Lauren class of diffuse samples, which have a poor prognosis.