NBDriver (NEIGHBORHOOD Driver) is a tool used to differentiate between driver and passenger mutations using features derived from the neighborhood sequences of somatic mutations.
Using missense mutation data from experimental assays, we build a binary classifier by extracting features from the neighborhood sequences of driver and passenger mutations. Our key results are three-fold. First, we use generative models to derive the distances between the underlying probability estimates of the neighborhood sequences for the two classes of mutations. Then, we build robust classification models using repeated cross-validation experiments to derive the median values of the metrics designed to estimate the classification performances. Finally, we demonstrate our models’ ability to predict unseen coding mutations from independent test datasets derived from large mutational databases.
The Brown et al. dataset was used as training data for our analysis. Raw nucleotide sequences surrounding the mutations published in this study were extracted from the reference genome build GRCH37. Then, seven feature representations, namely, TFIDF Vectorizer (k-mer sizes 2,3 and 4), Count Vectorizer (k-mer sizes 2,3 and 4) and One-hot encoding were used to convert the string-based features to numerical formats. This was followed by estimating the underlying probability distributions using kernel density estimation and repeated cross-validation experiments using Random Forests, KDE classifer and Extra Trees classifier. The final model (NBDriver) was obtained using a training set derived after removing all overalapping mutations between Brown et al. and an independent test set published by Martelotto et al. Subsequent validation with four separate independent validation sets containing pathogenic data from landamrk studies was also performed to judge the ability of NBDriver in predicting unseen test instances. The overall workflow is summarized below.
Training data was derived from a study by Brown et al., where they published mutation data from experimental assays labelled as drivers/passengers.
Brown AL, Li M, Goncearenco A, Panchenko AR (2019) Finding driver mutations in cancer: Elucidating the role of background mutational processes. PLOS Computational Biology 15(4): e1006981. https://doi.org/10.1371/journal.pcbi.1006981
Independent test dataset from a benchamrking study by Martelotto et al. consisted of 989 labelled driver and passenger mutations.
Martelotto, L.G., Ng, C.K., De Filippo, M.R. et al. Benchmarking mutation effect prediction algorithms using functionally validated cancer-related missense mutations. Genome Biol 15, 484 (2014). https://doi.org/10.1186/s13059-014-0484-1
scikit-learn - 0.22.1
pandas - 0.25.3
numpy - 1.18.5
imblearn - 0.5.0
ggplot2 - 3.3.2
reshape2 - 1.4.4
stringr - 1.4.0
tidyr - 1.1.2
readr - 1.4.0
caret - 6.0.86
Banerjee, S.; Raman, K.; Ravindran, B. Sequence Neighborhoods Enable Reliable Prediction of Pathogenic Mutations in Cancer Genomes. Cancers 2021, 13, 2366. https://doi.org/10.3390/cancers1310236
- Initiative for Biological Systems Engineering
- Robert Bosch Centre for Data Science and Artificial Intelligence (RBCDSAI)
