1. Milko Krachunov2
, Ivan Popov1
, Valeria Simeonova2
, Irena Avdjieva1
,
Paweł Szczęsny3
, Urszula Zelenkiewicz3
, Piotr Zelenkiewicz3
,
Dimitar Vassilev1
1
Bioinforomatics group, AgroBioInstitute, Bulgaria
2
Faculty of mathematics and informatics; Sofia University “St. Kliment Ohridski”, Bulgaria
3
Institute of Biochemistry and Biophysics, Polish Academy of Sciences, Warsaw, Poland
Detection and correction of errors in
metagenomic 16S RNA parallel sequencing

2. NGS errors – common problems
 Introduced errors in the assembled reads due to
imperfections both of biological and mathematical origin;
Impossibility to re-sequence the same sample again in
metagenomic studies ;
Tendency the error rate to increase in every step of the
process;
No easy way to differentiate between “sequencing error” and
“rare variant”;
Many existing methods and algorithms concerning different
aspects of the problem but no unified solutions are available;
Large amounts of data are difficult to process with common
software.

3. Significance of 16S RNA sequencing
Highly conserved between different species of bacteria and
archaea;
Sequence analysis is done with universal PCR primers;
Contains hypervariable regions that can provide species-
specific signature sequences;
Suitable for phylogenetic studies;
Suitable for metagenomic studies.

4. General approach in metagenomic biodiversity studies
454 Sequencing
Filtering / Denoising
Multiple alignment
Distance matrix
ОTU clusters with abundance count

5. Our approach:

6. A. Raw data characteristics and processing
Two separate runs of metagenomic 16S RNA fragments,
sequenced with 454 platform and converted in FASTA format:
run 02 – 46429 short reads
run 04 – 41386 short reads
Our task – extract, denoise and correct only the quality
reads.

7. Raw data length histogram
Run 02 Run 04

8. B. Correction with SHREC

9. C. Correction with our method:

10. Classification and performance evaluation
ClaMS parameters:
Distance cut-off: 0,05
Signature type: DBC
k-mer length: 3
Existing taxonomy: 4th Level

11. Aim of the method – idea outline
To deal with the heterogeneous nature of the data, similar or
related sequences are considered more important in the error
evaluation
The naïve approach: If a base is less common than the
sequencer error rate, assume it’s likely an error and replace
with the most common base
Our modification: Calculate the occurrence of the base in
reads that are similar in the given region – assign them bigger
weights or use them exclusively

12. Progress so far
Calculate occurrence rates of every base in reads that are
identical to the evaluated read in a window with radius of n
bases
 Preliminary results: The first basic implementation leads to
an increase in the number of OTUs found with ClaMS
Under development
 Good choice(s) of approach for alignment of the reads
 Empirical evaluation of the parameters
 Comparative evaluation of the variants of the approach

13. Software used in this project:
Python: http://www.python.org/
Cython: http://cython.org/
MEGA (Molecular Evolutionary Genetics Analysis):
http://www.megasoftware.net/
Muscle: http://www.drive5.com/muscle/
SHREC (SHort Read Error Correction method):
http://ww2.cs.mu.oz.au/~schroder/shrec_www/
ClaMS (Classifier for Metagenomic Sequences): http://clams.jgi-
psf.org/
NINJA (modified): http://nimbletwist.com/software/ninja/index.html
R-package: http://www.r-project.org/

14. milko@3mhz.net
Thank you