Bioinformatics Seminar
Time:
Venue: Na
25 February 2020
NaImproved variant effect prediction through application of machine learning to deep mutational scanning data
James FuWEHI Bioinformatics
With the rise of high-throughput sequencing technology ; missense mutations can now easily be identified in patients using whole genome and whole exome sequencing. However ; our understanding of the effect of such variants on protein function is limited. In order to address this problem ; approaches like in silico variant effect predictors as well as large-scale functional genomic experiments like deep mutational scanning (DMS) have been developed. Traditional variant effect predictors use evolutionary and structural features to predict the impact of variants. DMS experiments measure the variant effect landscape more directly by using biological assays and high throughput sequencing. For the first part of my PhD project ; I re-implemented two previously published variant effect predictors that were trained using DMS data as well as evolutionary and structural features. I also developed two new methods derived from recommender systems that leverage these data. I applied the predictors to imputing existing DMS data and also developed an ensemble imputation model which shows higher imputation accuracy than any of the individual methods.;;;;