Clinically and functionally uncharacterized variants in oncogenes are regularly encountered during tumor molecular profiling. Additional approaches, such as the use of computational algorithms, can be applied to determine their potential clinical relevance.
Different computational algorithms were combined to develop rules which can be used to predict oncogenic or neutral status with high confidence based on variants consistently annotated as oncogenic or neutral in JAX and OncoKB databases. The dataset including mutations in 41 oncogenes, which mutations may be associated with targeted therapy sensitivity, was used as a training set. MSK-IMPACT dataset was used for evaluation of practical use of defined algorithms.
A total of 2785 variants were included in the training dataset (754 consistent oncogenic and 184 consistent neutral). The best results in predicting oncogenic variants were achieved using a combination (in silico signature) of CHASMplus, VEST4, CADD, and PROVEAN, resulting in 36,34% sensitivity at 100% specificity. To validate a practical use of developed in silico signatures samples from the MSK-IMPACT study containing mutations in 41 oncogenes were used. The resulting dataset contained 5391 samples with a total of 9165 single nucleotide missense variant and 4121 unique variants. 5361 variants were annotated as oncogenic or likely oncogenic in OncoKB, 150 - as Neutral/Likely neutral/Inconclusive/Resistance, and 3654 were not annotated. In silico signature allowed to predict 4731 (51.6%) variants as oncogenic. Among them 4319 (91.3%) were annotated as oncogenic/likely oncogenic, 11 (0.2%) - as Neutral/Likely neutral/Inconclusive/Resistance, and 401 (8.5%) variants were not annotated in OncoKB. Dataset analysis by sample showed that 3.1% of patients (n=166) had not any known oncogenic variant but had at least one predicted variant. And 6.5% of patients (n=350) had at least one predicted but not annotated oncogenic variant.
The developed approach may be useful for identifying a subgroup of patients with potentially oncogenic variants who may be priority candidates for inclusion in clinical trials of appropriate targeted therapy.
The authors.
Has not received any funding.
All authors have declared no conflicts of interest.