Table 3.
Comparison of Overall Alignment Rate of Fiona versus RACER (with and without Athena’s tuning).
| — | Correlation to Alignment | Comparison with FIONA | Assembly quality | Runtime Improvement | |||||
|---|---|---|---|---|---|---|---|---|---|
| Dataset | Correlation (N-Gram) | Correlation (RNN) | Fiona + Bowtie2 (Alignment Rate) | RACER w/o Athena + Bowtie2 (Alignment Rate) | RACER w/Athena + Bowtie2 (Alignment Rate) | NG50 of Velvet w/o EC | NG50 of Velvet w/(Racer + Athena) | Athena | Bowtie2 |
| D1 | −0.977 | −0.938 | 99.25% | 85.01% | 99.26% | 3019 | 6827 (2.26X) | 1 m 38 s | 10 m 5 s |
| D2 | −0.981 | −0.969 | 73.75% | 58.66% | 81.15% | 47 | 2164 (46X) | 49 s | 3 m 53 s |
| D3 | −0.982 | −0.968 | 83.12% | 80.79% | 84.11% | 1042 | 4164 (4X) | 1 m 39 s | 7 m 50 s |
| D4 | −0.946 | −0.930 | 95.33% | 93.86% | 95.33% | 118 | 858 (7.27X) | 52 s | 3 m 8 s |
| D5 | −0.970 | −0.962 | 92.34% | 90.91% | 92.29% | 186 | 2799 (15X) | 1 m 40 s | 9 m 42 s |
| D6 | −0.944 | −0.979 | 87.43% | 85.76% | 86.84% | 1098 | 1237 (1.12X) | 6 m 40 s | 1 h 42 m |
| D7 | −0.723 | −0.862 | NA | 17.17% | 17.55% | 723 | 754 (1.04X) | 16 m | 71 m |
RACER requires the user to enter a value for the “Genome Length”, which has no default value. Therefore, “RACER w/o Athena” is RACER operating with a fixed Genome Length of 1M. Columns 5 & 6 demonstrate the strong anti-correlation values between Perplexity and Alignment Rate. The last two columns show the assembly quality (in terms of NG50) before and after correction by RACER, tuned with Athena. Improvements in NG50 are shown between parentheses, while NGA50 and the amount of assembly errors metrics showed similar improvements and hence omitted. We also show the search time comparison for estimating the perplexity metric with Athena (N-gram) for a point in search space vs. estimating overall alignment rate with Bowtie2.