Some puzzling parameters in defuse

[swordsman]

I am focusing on some constantly incorrectly rejected fusion genes. After I replace the reference and annotation data according to the wiki page of defuse on the Internet, there are still some parameters I cannot figure out. The parameters are:
// Filtering parameters
max_dist_pos = 600
num_dist_genes = 500
split_min_anchor = 4 // what is the difference between split_count_threshold and it?
splice_bias = 10

// Position density when calculating covariance
covariance_sampling_density = 0.01
// Number of regions for each breakpoint sequence job in split
regions_per_job = 20

Thanks,
Daqing

[Andrew]

Here are the explanations of the parameters:

split_min_anchor: minimum number of nt that must align to one side or the other of a breakpoint for the split read to be valid

splice_bias: used to calculate a genomic position given a transcriptomic position. If the fusion boundary prediction is just past a splice junction, but in reality the fusion boundary is exactly at the splice junction, the wrong genomic position will be reported. Thus we subtract splice_bias, remap to the genome, and then add splice_bias to the remapped coordinate.

covariance_sampling_density: we attempt to account for non-independence of read fragmentation by calculating a covariance between fragment lengths of reads that span a particular positon.

max_dist_pos, num_dist_genes and regions_per_job are deprecated, perhaps you have an outdated config file.

If you have a public dataset with known fusions that defuse is not finding please let me know and I will try the dataset myself.

Also, there is another post which addresses false negatives, please see that post for some tips.