Recall (a) and precision (b) scores for simulated insertions.
Filtering is tested with three simulated read sets with different insert size distributions. As the results for our proposed filtering scheme differ based on the insert size distributions, we have separated and labeled them with the corresponding mean insert sizes, μ = 150, 1500, 3000. The other two are averaged over all the read sets. The dashed lines for Filter use only reads that are filtered in and the solid lines add all unmapped reads to the filtered read set if the coverage of the filtered reads is below the coverage threshold (here 25). The precision of using all unmapped reads is almost zero and is thus not visible in the graph.