You are on page 1of 2

Machine learning insights into the spatiotemporal assembly of microbiome

in a full-scale sequencing batch reactor wastewater treatment plant

Jonathan Wijaya

Keywords: Wastewater treatment, Sequencing batch reactor (SBR), Machine learning, Artificial intelligence,
Microbiome, 16S rRNA gene

1 Introduction bayesian network modelling with probabilistic


Microbiomes in conventional activated sludge mechanism also performed to complete the
process (CAS) has been explored over the years study. The bayesian network use tabu algorithm
for its functioning conditions and performance with 0.8 threshold for the strength variable.
efficiency. High throughput sequencing analysis Statistical analysis also conducted using
has been used to explore the diversity in PAleonological STatistic (PAST) software and
wastewater treatment plants (WWTPs) (Ju the figures were generated using R and python.
Zhang, 2015; Saunders et al., 2016; Wu et al.,
2019). Most previous studies are discussed the 3 Results and Discussion
importance of microbiomes data in CAS system
rather than the sequencing batch reactor (SBR)
process. Therefore, this study focused on the
microbiome interaction in the SBR process
along with the machine learning approach and
bayesian network modelling.

2 Materials/Methods
In this study, two WWTPs with different
operational characteristics are used (A2O, and Figure 1: Alpha dan Beta diversity from 2 types
SBR). Monthly samples were taken from both of WWTPs.
aerobic tanks in each WWTPs from February
2018 until July 2019. Triplicate samples from Figure 1A shows that the A2O and SBR
each process (anaerobic and aerobic) and processes can be differentiate by the ordination
influent in SBR were also taken in February analysis (NMDS). The stress value of < 0.3
2018. The DNA samples were extracted and shows that the NMDS plot could represent the
assessed for its quality and quantity using ordination analysis. The A2O and SBR process
manufacturer standards. Polymerase chain also significantly different (P < 0.05) using
reaction (PCR) amplification of 16S rRNA PERMANOVA. Figure 1B-E shows the alpha
genes and sequencing using MiSeq sequencer diversity indices from Chao, Ace, Shannon, and
were conducted on the DNA samples. The raw Inverse Simpson. Asterisk symbols shows the
pair-end sequencing results were pre-processes significantly difference between groups. SBR
using bioinformatics tool (MOTHUR) for processes has lower richness, but higher
monitoring the quality of sequences and diversity.
removing the chimeric sequences. The
sequences were clustered to the taxonomy
alignment and operational taxonomy units
(OTUs) with > 97% nucleotide identity cut-off.
The results of MOTHUR generates table with
OTUs number, taxonomy, and counts of
sequences in each sample. Classification and
regression analysis using six different machine
learning (ML) modelling were constructed to
show the prediction performance of WW
operational characteristics (A2O, or SBR) and
environmental factor (water temperature). In
addition to the machine learning analysis, Figure 2: ML model classification performance
The core microbiome for ML modelling was
chosen based on the pareto law chart with the
occurrence more than 80% and relative
abundance more than 1%. Figure 2 shows the
results of ML prediction modelling in
classifying the A2O and SBR process. The
results show the high prediction accuracy from 4
different models (SVMRBF, SVML, LR, and
RF) with accuracy ranged from 93-96% and area
under the curve (AUC) from 0.98-0.99. The
confusion matrix in Figure 2C-F show the
detailed prediction from each model. The
highest prediction performance was achieved by
the SVMRBF and SVML, which were the
support vector machine algorithms.

Figure 4: Correlation between water temperature


and core 18 families.

Figure 3: Prediction impact of the core 11 OTUs


4 Conclusions
The machine learning and Bayesian network
in ML classification models.
modelling were used in this study to predict the
operational characteristics and environmental
The ML modeling using the feature weight
factors of WWTPs. It shows that Chloroflexi
analysis also shows the importance of each
plays an important role in SBR process.
variable in differentiating the SBR and A2O
process. Figure 3 shows that Chloroflexi,
Gordonia, and Sphingobacteriales as the major 5 References
Ju, F. and Zhang, T. 2015. Bacterial assembly and
microbes to differentiate the operational
temporal dynamics in activated sludge of a full-
characteristics. OTUs inside the Chloroflexi scale municipal wastewater treatment plant. Isme
phylum were then investigated to further J 9(3), 683-695.
analyze the impact of each OTUs for the Saunders, A.M., Albertsen, M., Vollertsen, J. and
operational characteristics. Figure 4E shows the Nielsen, P.H. 2016. The activated sludge
phylogenetic tree of TOP 8 OTUs in Chloroflexi ecosystem contains a core community of
where it shows that some OTUs in Chloroflexi abundant organisms. Isme J 10(1), 11-20.
could be enriched more in A2O or SBR process. Wu, L.W., Ning, D.L., Zhang, B., Li, Y., Zhang, P.,
Figure 4 shows the consistent result from Shan, X.Y., Zhang, Q.T., Brown, M.R., Li, Z.X.,
bayesian network and ML modelling for water Van Nostrand, J.D., Ling, F.Q., Xiao, N.J.,
temperature predictions. Zhang, Y., Vierheilig, J., Wells, G.F., Yang,
Y.F., Deng, Y., Tu, Q.C., Wang, A.J., Zhang, T.,
He, Z.L., Keller, J., Nielsen, P.H., Alvarez,
P.J.J., Criddle, C.S., Wagner, M., Tiedje, J.M.,
He, Q., Curtis, T.P., Stahl, D.A., Alvarez-Cohen,
L., Rittmann, B.E., Wen, X.H., Zhou, J.Z. and
Consortium, G.W.M. 2019b. Global diversity
and biogeography of bacterial communities in
wastewater treatment plants (vol 4, pg 1183,
2019). Nat Microbiol 4(12), 2579-2579.

You might also like