Patient-specific driver gene prediction and risk assessment through integrated network analysis of cancer omics profiles.
Bottom Line: Extensive in silico and in vitro validation helped establish OncoIMPACT's robustness, improved precision over competing approaches and verifiable patient and cell line specific predictions (2/2 and 6/7 true positives and negatives, respectively).In particular, we computationally predicted and experimentally validated the gene TRIM24 as a putative novel amplified driver in a melanoma patient.We also provide the first demonstration that computationally derived driver mutation signatures can be overall superior to single gene and gene expression based signatures in enabling patient stratification and prognostication.
Affiliation: Computational and Systems Biology, Genome Institute of Singapore, Singapore 138672, Singapore.Show MeSH
Related in: MedlinePlus
Mentions: While the identification of patient-specific driver genes is challenging, validating a methodology that identifies them is even more so, given the lack of gold standards (e.g. by their very definition patient-specific drivers are less likely to be in CGC). We attempted to verify OncoIMPACT's ability to call patient-specific drivers using three different approaches. First, we experimented with in silico data sets derived from real TCGA data sets by introducing random mutations to test our ability to discriminate them. These experiments highlight that OncoIMPACT shows a high-degree of tolerance to the introduction of decoy mutations, and can robustly accommodate up to 10% of erroneous mutation data (e.g. due to sequencing or variant-calling errors) (Figure 3a, Supplementary Figure S6). In doing so, it is able to control the false positive rate (FPR) to be generally less than 10% (median FPR < 5% for 2.5% decoy mutations), suggesting that a majority of patient-specific driver predictions are likely to be true positives. To further validate the consistency of patient-specific driver gene predictions, we experimented with learning phenotype genes from a random subset of samples, with prediction on unselected samples (cross-validation). Our results show good predictive stability for all drivers and good predictive recovery for both common (>5% frequency) and rare drivers (<5% frequency) (Supplementary Figure S7), further confirming OncoIMPACT's robustness for patient-specific driver gene prediction.
Affiliation: Computational and Systems Biology, Genome Institute of Singapore, Singapore 138672, Singapore.