Limits...
Significant loss of sensitivity and specificity in the taxonomic classification occurs when short 16S rRNA gene sequences are used

View Article: PubMed Central - PubMed

ABSTRACT

The classification performance of Kraken was evaluated in terms of sensitivity and specificity when using short and long 16S rRNA sequences. A total of 440,738 sequences from bacteria with complete taxonomic classifications were downloaded from the high quality ribosomal RNA database SILVA. Amplicons produced (86,371 sequences; 1450 bp) by virtual PCR with primers covering the V1–V9 region of the 16S-rRNA gene were used as reference. Virtual PCŔs of internal fragments V3–V4, V4–V5 and V3–V5 were performed. A total of 81,523, 82,334 and 82,998 amplicons were obtained for regions V3–V4, V4–V5 and V3–V5 respectively. Differences in depth of taxonomic classification were detected among the internal fragments. For instance, sensitivity and specificity of sequences classified up to subspecies level were higher when the largest internal fraction (V3–V5) was used (54.0 and 74.6% respectively), compared to V3–V4 (45.1 and 66.7%) and V4–V5 (41.8 and 64.6%) fragments. Similar pattern was detected for sequences classified up to more superficial taxonomic categories (i.e. family, order, class…). Results also demonstrate that internal fragments lost specificity and some could be misclassified at the deepest taxonomic levels (i.e. species or subspecies). It is concluded that the larger V3–V5 fragment could be considered for massive high throughput sequencing reducing the loss of sensitivity and sensibility.

No MeSH data available.


Classification output obtained with the internal fractions (V3–V4, V4–V5, V3–V5) respect to the complete 16S rRNA gene sequences (V1–V9). Proportion of sequences with same classification results using either the complete sequence or the internal fragment sequence are labeled as “Equal”; whereas sequences of internal fractions with similar classification but to more superficial level are represented as “Less”, and sequences with different taxonomic classification compared to the respective complete sequence (V1–V9) are named as “Low”.
© Copyright Policy - CC BY-NC-ND
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5037269&req=5

fig0015: Classification output obtained with the internal fractions (V3–V4, V4–V5, V3–V5) respect to the complete 16S rRNA gene sequences (V1–V9). Proportion of sequences with same classification results using either the complete sequence or the internal fragment sequence are labeled as “Equal”; whereas sequences of internal fractions with similar classification but to more superficial level are represented as “Less”, and sequences with different taxonomic classification compared to the respective complete sequence (V1–V9) are named as “Low”.

Mentions: For instance, 41,685 unilarge sequences (V1–V9) resulted to be classified up to Subspecies level; however, when used as internal V3–V5 fragments, only 22,520 (55.6%) sequences received the same classification than the unilarge fragments (subspecies), whereas 10,308 (25.5%) were similarly classified but to less deep taxonomic levels (species, or genus, or family, and so on) (Fig. 3). The rest of the sequences received a different (and erroneous) classification output; for instance, 7,669 (18.9%) sequences obtained after the amplification of the V3–V5 region were labeled as a different organism (other species, or genus, or family, and so on); 11 of these sequences were inclusively classified up to Subspecies, but to a different specimen.


Significant loss of sensitivity and specificity in the taxonomic classification occurs when short 16S rRNA gene sequences are used
Classification output obtained with the internal fractions (V3–V4, V4–V5, V3–V5) respect to the complete 16S rRNA gene sequences (V1–V9). Proportion of sequences with same classification results using either the complete sequence or the internal fragment sequence are labeled as “Equal”; whereas sequences of internal fractions with similar classification but to more superficial level are represented as “Less”, and sequences with different taxonomic classification compared to the respective complete sequence (V1–V9) are named as “Low”.
© Copyright Policy - CC BY-NC-ND
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5037269&req=5

fig0015: Classification output obtained with the internal fractions (V3–V4, V4–V5, V3–V5) respect to the complete 16S rRNA gene sequences (V1–V9). Proportion of sequences with same classification results using either the complete sequence or the internal fragment sequence are labeled as “Equal”; whereas sequences of internal fractions with similar classification but to more superficial level are represented as “Less”, and sequences with different taxonomic classification compared to the respective complete sequence (V1–V9) are named as “Low”.
Mentions: For instance, 41,685 unilarge sequences (V1–V9) resulted to be classified up to Subspecies level; however, when used as internal V3–V5 fragments, only 22,520 (55.6%) sequences received the same classification than the unilarge fragments (subspecies), whereas 10,308 (25.5%) were similarly classified but to less deep taxonomic levels (species, or genus, or family, and so on) (Fig. 3). The rest of the sequences received a different (and erroneous) classification output; for instance, 7,669 (18.9%) sequences obtained after the amplification of the V3–V5 region were labeled as a different organism (other species, or genus, or family, and so on); 11 of these sequences were inclusively classified up to Subspecies, but to a different specimen.

View Article: PubMed Central - PubMed

ABSTRACT

The classification performance of Kraken was evaluated in terms of sensitivity and specificity when using short and long 16S rRNA sequences. A total of 440,738 sequences from bacteria with complete taxonomic classifications were downloaded from the high quality ribosomal RNA database SILVA. Amplicons produced (86,371 sequences; 1450 bp) by virtual PCR with primers covering the V1–V9 region of the 16S-rRNA gene were used as reference. Virtual PCŔs of internal fragments V3–V4, V4–V5 and V3–V5 were performed. A total of 81,523, 82,334 and 82,998 amplicons were obtained for regions V3–V4, V4–V5 and V3–V5 respectively. Differences in depth of taxonomic classification were detected among the internal fragments. For instance, sensitivity and specificity of sequences classified up to subspecies level were higher when the largest internal fraction (V3–V5) was used (54.0 and 74.6% respectively), compared to V3–V4 (45.1 and 66.7%) and V4–V5 (41.8 and 64.6%) fragments. Similar pattern was detected for sequences classified up to more superficial taxonomic categories (i.e. family, order, class…). Results also demonstrate that internal fragments lost specificity and some could be misclassified at the deepest taxonomic levels (i.e. species or subspecies). It is concluded that the larger V3–V5 fragment could be considered for massive high throughput sequencing reducing the loss of sensitivity and sensibility.

No MeSH data available.