Models to predict Mutational Burden in Cancer

CRISTINA MARINO BUSLJE; FERNANDO ORTI; ELIZABETH MARTINEZ PEREZ

Mar del Plata

Congreso; 9th CAB2C; 2018

ASOCIACIÓN ARGENTINA DE BIOINFORMÁTICA y BIOLOGÍA COMPUTACIONAL

AbstractBackgroundImmune checkpoint inhibitors treatments is a promising therapy as it probed effective in a subset of patients. It has been observed that a high number of non synonymous mutations (NSM), also known as mutational burden (MB), correlates with a good response to checkpoint inhibitors in melanoma and lung. NSM can generate neoantigens, which, in turn, can be recognized by the immune system, triggering an anticancer immune response. While treating cancer with immunotherapy can be highly effective, only some patients respond to this treatment, so, there is a great interest to discriminate patients most likely to benefit from this therapy. The goal of that work is to provide the subset of genes or exons needed to be sequenced in order to predict the MB.MethodsWe have worked with COSMIC whole genome v84 (25533 samples, whole genome/exome sequenced). After filtering, the working data involve 24726 samples from 42 cancer types. We obtained predictor models of MB for each cancer type, based on genes and exons. Predictors were made with linear models using forward step selecting strategy. Predictors were tested with internal (3 times 10 fold cross-validation) and external data ( published studies on patient cohorts).ResultsWe obtained several models for each cancer type. The most predictive models have similar explained variance (R2 metric value). While exon based models benefit form sequencing less base pairs (bp), gene based models have smaller error (RMSE metric). Good models (with R2 > 0.8) were achieved in 33 out of 42 cancer types, regular models (between 0.6 < R2 < 0.8) in 5 and bad models (R2 < 0.6) in 4. ConclusionsModels to predict MB in 42 cancer types were made and tested. The models provide the selection set of genes/exons needed to predict the MB. This is particularly important to reduce sequence cost, as whole genome/exome sequencing is very expensive. The sets of genes/exons obtained with the models indicate a number of bp to be sequenced even smaller than current commercial panels. Patients with a high MB will be the more likely to benefit from immunotherapy treatment.