Genetic Variant Positioning Simulation

MOATsim A somatic variant simulator Given a set of input variants, shuffle to new locations, taking genome structure into account MOAT-v s permutation step without the p-value calculations 1

MOATsim Read genome coordinates into memory Divide into bins of user-specified size Subtract blacklist regions Use bigWigAverageOverBed to derive covariate values for each bin Generate covariate matrix covar 1 covar 2 covar 3 bin 1 bin 2 bin 3 bin 4 bin 5 2

Blacklist filtering Relevant variables: bin width: The width that represents the local genome context. Variants are shuffled to new locations within their containing bin. min bin width: In the event that an especially small bin is formed, either due to a chromosome end or subtraction of a blacklist region, merge the bin with the nearest full size bin if it s below this width. bin width bin width 3

Blacklist filtering Excludes low mappability regions Results in bins smaller than bin_width Possibly too small to offer many new variant locations Hence, guarantee a minimum bin width min_bin_width user parameter Typically set to half the size of bin_width truncated bin_width 4

Blacklist filtering Bins smaller than min_bin_width are merged with an adjacent neighbor If no adjacent neighbor is available, remove the bin bin_width merged 5

Covariate matrix: row clustering Goal: Find similar bins (i.e. similar covariate vectors) and treat as single block Currently using k-means clustering Found that the optimal cluster number k was around 30 using MutSig covariate file (defined over exome only) Within-sum-of-squares prone to stochastic variation beyond this point 50000 40000 Within groups sum of squares 30000 20000 Rough location for beginning of Stochastic variation in within-ss 10000 6 0 10 20 30 40 50 Number of Clusters

Variant Placement Step Start with many bins Mark which have similar covariate vectors Given input variants Shuffle to new locations within bin cluster 7