Skip to main content


Fig. 5 | Biology Direct

Fig. 5

From: xHMMER3x2: Utilizing HMMER3’s speed and HMMER2’s sensitivity and specificity in the glocal alignment mode for improved large-scale protein domain annotation

Fig. 5

Domain-detection sensitivity and speed differences of HMMER2 and xHMMER3x2 against HMMER3. The leftmost boxplot represents the baseline difference where the average speed of HMMER2 is about 222× slower than HMMER3. However, the sensitivity of HMMER2 is 4.3% (100% versus 95.7%) higher than that of HMMER3. The rightmost boxplot represents the fastest xHMMER3x2 by running all HMMER3 models where its sensitivity is equal to that of HMMER3 at 95.7%. The trade-offs between computational speed and domain-detection sensitivity is given by the middle boxplots. As N (i.e., the number of default HMMER2 models) decreases from 3502,1500,1000 to 500 (see second to fifth boxplots from right), xHMMER3x2’s sensitivity decreases from 99.8, 99.3, 99.1 to 98.6% respectively. In contrast, its speed increases accordingly from 32×, 17×, 12× to 6.2× (M h2/(M h3 + K h3) = 100.81) slower than HMMER3. Alternatively, this translates to 7×, 13×, 18.5× and 36× faster than HMMER2 respectively. The best compromise is achieved via the top 1000 most sensitive HMMER2 domains with the remaining as HMMER3 models, to attain a domain-sensitivity of 99.1% and a speed of 12× slower than HMMER3 but 18.5× faster than HMMER2

Back to article page