Table 2. Results obtained by several tools with several parameter configurations on two human chromosome 1 diploid datasets.
| Tool | Precision (%) | Recall (%) | Time | Memory (GB) |
|---|---|---|---|---|
| discoSnpb0 | 96.98 | 71.99 | 4h24 | 5 |
| cortex | 97.22 | 69.70 | 10h09 | 105 |
| bubbleparsed0 | 95.78 | 72.71 | 11h24 | 105 |
| discoSnpb1 | 92.33 | 79.22 | 4h44 | 5 |
| bubbleparsed1 | 91.66 | 76.60 | 11h00 | 105 |
| hybrid strategy | 96.18 | 72.86 | 10h34 | 54 |
The upper part of the table shows results obtained by de novo tools with stringent parameters (no branching allowed in bubbles). Below are the results obtained with the same tools but with more sensitive parameters, namely allowing for some branchings in bubbles. The bottom line shows results obtained with a hybrid strategy (SOAPdenovo2 + Bowtie2 + GATK), filtering out low covered SNPs (with the same parameters as those applied for all the other tools, by removing SNPs whose both alleles have coverage below four). As presented in Supplementary Additional File 1, results are worse without this filter.