SZTE  Doktori Értekezések Repozitórium (SZTE Repository of Dissertations)

Oltean-Polanek Evelin

English

Micronutrient Supplementation Among Hungarian Pregnant Women With a Special Focus on Vitamin D

Kiss Aletta Kata

The role of SPCA2 in store-independent STIM1-ORAI1 activation and the regulation of basal CFTR activity in epithelial secretion

Nagy Viktória

Use of Advanced Transthoracic Echocardiography in Special Cardiac Conditions

Takács Bertalan Vilmos

Development of a Novel 
Metagenomic Classification Method 
PhD thesis booklet 
Bertalan Vilmos Takács 
Supervisor: 
Dr. Lajos Haracska 
DOCTORAL SCHOOL OF BIOLOGY  
FACULTY OF SCIENCE AND INFORMATICS 
UNIVERSITY OF SZEGED 
INSTITUE OF GENETICS 
HUN-REN BIOLOGICAL RESEARCH CENTRE 
2025 
Magyar nyelvű összefoglaló 
Az emberi mikrobiom az emberi szervezettel együtt élő 
mikroorganizmusok összessége, amelyek pozitív és 
negatív módon egyaránt befolyásolhatják az egészséget. 
A mikrobiális összetevők azonosításához olyan 
algoritmusokra van szükség, amelyek nagy pontossággal 
határozzák meg a minták összetételét, minimalizálják a 
hamis pozitív találatokat, és egyben gyorsak, valamint 
skálázhatók is. 
Doktori munkám célja egy új metagenomikai azonosító 
algoritmus, a Novel Alignment-based Biome Analysis 
Software+ (NABAS+) fejlesztés és összehasonlító 
tesztelése volt. A programot Java nyelven 
implementáltuk, kihasználva a BWA-MEM 
illesztőalgoritmus párhuzamosíthatóságát és a RefSeq 
adatbázis magas megbízhatóságú referencia-genomjait. 
A NABAS+ teljesítményét három típusú adatsoron 
vizsgáltuk: 
 (1) In silico: Saját fejlesztésű (30 minta) és a CAMI II “toy 
human gastrooral” adatsor (20 minta). A NABAS+ 
pontossága összemérhető, több esetben jobb volt, mint a 
széles körben használt osztályozóké (Kraken2, 
MetaPhlAn3, GOTTCHA). Egy minta (sample19) 
újragenerálása tovább növelte a találatok pontosságát. 
 (2) In vitro: Zymo Community Standard I és II (CSI, CSII) 
Illumina-szekvenált adatsorai. A NABAS+ pontosan 
detektálta az alacsony abundanciájú fajokat, és nem 
jelzett hamis pozitív találatot. 
 (3) Klinikai adatok: 20 humán bélmikrobiom-mintát 
elemeztünk; az algoritmus az adatsor 11 fertőzött 
mintájában helyesen azonosította a kórokozót, és nem 
mutatott hamis pozitív eredményt. 
Összefoglalva: kutatásom során egy új, teljes genom 
illesztésen alapuló metagenomikai azonosító szoftvert 
fejlesztettem és validáltam. A NABAS+ teljesítménye a 
legjobb jelenlegi algoritmusokkal összemérhető, 
alacsony abundanciájú fajokat és patogéneket is 
pontosan detektál, így nemcsak kutatási, hanem klinikai 
környezetben is megbízhatóan alkalmazható. Publikus 
elérhetősége elősegítheti a metagenomikai vizsgálatok 
pontosságának, reprodukálhatóságának és szélesebb 
tudományos felhasználásának növelését. 
Summary in English 
Introduction and Background 
The human microbiome encompasses the complex 
community of microorganisms living in and on the human 
body, including bacteria, archaea, viruses, and eukaryotic 
microbes. These organisms can have both beneficial and 
detrimental effects on human health, influencing 
metabolism, immune function, and disease susceptibility. 
The growing use of next-generation sequencing (NGS) 
technologies has made it possible to study microbiomes 
with unprecedented depth, yet accurate taxonomic 
classification remains a major methodological challenge.  
This doctoral work focuses on classifying bacteria in 
shotgun sequenced metagenomic samples coming from 
the human microbiome. Although a wide range of 
algorithms available for this task, relying on different 
algorithmic principles, they are prone to produce 
discordant results when applied to the same samples. 
Reliable and reproducible species-level identification is 
particularly important when metagenomics is applied in 
clinical diagnostics, for example in detecting bacterial 
pathogens in stool samples. Misclassification—especially 
false positives—may influence treatment decisions. 
Despite significant progress in computational 
metagenomics, there is still no consensus on a “gold-
standard” classifier, nor on universally accepted 
threshold settings for post-classification filtering. 
For these reasons, we developed NABAS+ (Novel 
Alignment-Based Biome Analysis Software+), an 
alignment-based metagenomic classifier designed to 
minimize false positives while maintaining high accuracy 
across various sample types. NABAS+ relies on highly 
curated reference genomes, a modular database 
architecture, and the BWA-MEM alignment algorithm to 
provide robust taxonomic classification with minimal 
ambiguity. 
Research Objectives 
During my doctoral work, I set the following aims: 
 To develop a new metagenomic classification 
algorithm, optimized for precision, reproducibility, 
and low false-positive rates. 
 To construct in silico benchmark datasets suitable 
for evaluating classifier performance under 
controlled conditions. 
 To compare NABAS+ with widely used 
metagenomic classifiers on multiple dataset types, 
including in silico data, standardized mock 
communities, and real clinical samples. 
 To assess the suitability of NABAS+ for clinical 
diagnostics by testing its pathogen-identification 
accuracy on human stool metagenomic samples. 
Methods 
NABAS+ was implemented in Java and designed around 
the principle of using high-quality reference data. Only 
“representative” genomes from the RefSeq database 
were included, ensuring that the classifier relied on the 
most reliable assemblies available. Selecting a single 
genome per species reduced redundancy in the 
reference set and allowed NABAS+ to identify species 
with fewer ambiguous alignments. To make updates 
straightforward and memory usage manageable, the 
reference database was divided into modular chunks of 
approximately 8×10⁹ bases, each indexed separately 
with BWA-MEM. This modular structure not only reduced 
RAM requirements during alignment but also allowed 
new genomes to be added without rebuilding the entire 
database. 
Taxonomic assignment in NABAS+ is based on the 
alignment of paired-end Illumina reads (2×151 bp) to the 
reference genomes using the BWA-MEM algorithm. After 
alignment, reads undergo strict filtering based on edit 
distance, alignment quality, and minimum read counts. In 
addition, actual genome coverage and hypothetical 
genome coverage are computed, and genomes with 
disproportionately high coverage are excluded using a 
threshold ratio of 3.5. By combining curated reference 
genomes with strict alignment thresholds, NABAS+ aims 
to minimize the rate of false-positive identifications while 
retaining the ability to detect low-abundance species. 
To evaluate the performance of NABAS+, I used three 
types of datasets. First, I constructed six in silico mock 
communities representing different environments, 
including gut, oral, skin, and environmental microbiomes. 
These communities comprised 212 species in total and 
modelled samples of varying read depths, allowing the 
assessment of classification accuracy under ideal 
conditions. Additionally, I employed the CAMI II "toy 
human gastrooral" dataset, a well-established 
benchmark dataset containing samples with realistic read 
distributions and error profiles. 
Second, I tested NABAS+ on two in vitro datasets: the 
ZymoBIOMICS Microbial Community Standards I and II. 
These standardized communities contain eight bacterial 
species and two fungi at known relative abundances, 
making them ideal for evaluating classifier accuracy using 
real sequencing data. 
Finally, I examined the performance of NABAS+ on 20 
clinical human stool samples from Angel et al. (2025), 11 
of which were confirmed to contain bacterial pathogens 
by PCR or culture-based methods. This dataset allowed 
me to determine whether NABAS+ can reliably detect 
pathogens in real clinical metagenomes. 
Results 
Performance on in silico mock communities 
In the in silico experiments, NABAS+ consistently 
identified the correct number of species across all mock 
communities. Its performance was not influenced by the 
origin or complexity of the community, whether it 
represented an environmental microbiome or human-
associated microbiota such as gut or oral samples. We 
observed that the following, commonly used algorithms 
produced the result most closely to NABAS: MetaPhlAn3, 
Kraken2, and GOTTCHA. This result highlights the 
robustness of NABAS+ when applied to in silico 
metagenomic data. 
Sensitivity to sequencing depth 
To evaluate how sequencing coverage affects 
classification accuracy, I compared the performance of 
NABAS+ and other tools across read depths ranging from 
5×10⁵ to 1×10⁷. We observed substantial differences 
between classifiers in their sensitivity to increasing read 
numbers. DIAMOND and Kaiju were the most affected, 
showing pronounced increases in detected species with 
higher coverage, while Centrifuge, Bracken, and CLARK 
exhibited more moderate changes. By contrast, NABAS+, 
MetaPhlAn3, and GOTTCHA produced highly stable 
results, indicating that their performance was largely 
independent of sequencing depth. NABAS+ therefore 
maintained both accuracy and consistency across 
datasets modeling both deeply and shallowly 
sequenced samples.  
F1-score analysis 
To further quantify classifier performance, I calculated F1 
scores across all in silico datasets. NABAS+ showed 
consistently high F1 values, often matching or surpassing 
the most widely used tools. Kraken2, although strong in 
recall, frequently produced numerous false positives that 
reduced its precision. In contrast, NABAS+ achieved a 
balance between sensitivity and specificity that resulted 
in stable, high F1 performance across all tested 
communities. 
Evaluation with the CAMI II gastrooral dataset 
Using the CAMI II benchmark, I observed that 
MetaPhlAn3 performed best overall, followed closely by 
NABAS+ and GOTTCHA. Kraken2 again exhibited strong 
recall but poor precision. One sample, sample19, 
produced unusually poor results across all classifiers; 
subsequent investigation revealed that this sample had 
been generated from outdated genome assemblies. After 
regenerating the sample using the most recent RefSeq 
representative genomes, all classifiers showed improved 
accuracy. This finding illustrates that classifier 
performance is determined not only by the underlying 
algorithm but also by the quality and currency of the 
reference database. 
Results on Zymo mock communities 
On the Zymo community standards, NABAS+ achieved 
excellent results. In both the equal-abundance and 
logarithmic-abundance datasets, NABAS+ detected all 
organisms present in the sample and did not report any 
false positives. It also produced the lowest Bray–Curtis 
distance to the true community composition, indicating 
that its relative abundance estimates were more accurate 
than those of other classifiers. Notably, NABAS+ 
detected Enterococcus faecalis at an abundance as low 
as 0.00089%, demonstrating exceptional sensitivity for 
low-abundance species. 
Clinical validation 
In the clinical dataset of 20 human stool samples, 11 
contained bacterial pathogens confirmed by PCR or 
culture. NABAS+ correctly identified all pathogens in the 
positive samples and did not produce any false-positive 
results in the negative samples. This perfect concordance 
with laboratory diagnostics demonstrates that NABAS+ is 
suitable for clinical application and can reliably detect 
pathogens in complex human microbiomes.  
Summary and Scientific Contributions 
In this dissertation, I developed and validated NABAS+, 
a new alignment-based metagenomic classifier designed 
to minimize false positives and deliver reproducible, high-
precision species identifications. I demonstrated that 
NABAS+ performs comparably to or better than widely 
used classifiers such as Kraken2, MetaPhlAn3, and 
GOTTCHA across in silico, in vitro, and clinical datasets. 
NABAS+ showed strong resilience to sequencing depth, 
excellent accuracy on standardized reference 
communities, and perfect pathogen identification in real 
clinical samples. 
The scientific contributions of this work include the 
creation of a modular, easily updateable database 
structure; the empirical determination of optimal 
alignment filtering thresholds; the demonstration that 
database quality has a major impact on classification 
performance; and the application of NABAS+ to clinical 
metagenomic diagnostics. Together, these findings show 
that NABAS+ is a reliable and versatile tool that can 
enhance both research and clinical microbiome analysis. 
List of Publications 
MTMT ID: 10074085 
Publications used in this thesis 
1. Advancing metagenomic classification with 
NABAS+: a novel alignment-based approach 
Takács et al. 
NAR Genomics and Bioinformatics, Volume 
7, Issue 3, September 2025 
Impact Factor: 5.0 
2. Shotgun Analysis of Gut Microbiota with 
Body Composition and Lipid Characteristics 
in Crohn’s Disease  
Bacsur et al. 
Biomedicines 2024, 12(9), 2100; 
Impact factor: 3.9 
Other publications 
1. Prolonged activity of the transposase helper 
may raise safety concerns during DNA 
transposon-based gene therapy 
Imre et al. 
Molecular Therapy Methods & Clinical  
 Development, Volume 29, 145 – 159 
Impact Factor: 4.7 
2. A Comprehensive Evaluation of the 
Performance of Prediction Algorithms on 
Clinically Relevant Missense Variants 
Qorri et al. 
International Journal of Molecular Sciences. 
2022; 23(14):7946 
Impact Factor: 4.9 
3. Distinct Gut Microbiota Profiles in 
Unruptured and Ruptured Intracranial 
Aneurysms: Focus on Butyrate-Producing 
Bacteria 
Csécsei et al. 
Journal of Clinical Medicine. 2025; 
14(10):3488 
Impact Factor: 3.0 
Conference talks 
1. Comparative Analysis of the Microbial 
Composition of Different Tumor Tissues 
Talk at the Microbiota and Cancer Immunity 
Conference, Taipei, Taiwan, 2024 
2. The devil is in the details: Eliminating false 
metagenomic classification with a novel 
algorithm 
Talk at the 61st Annual Meeting of the 
Hungarian Society of Laboratory 
Diagnostics, Budapest, Hungary, 2023 
3. Understanding the effects of COVID-19 on 
the microbiome using bioinformatics and 
machine learning 
Talk at the 5th National Conference of Young 
Biotechnologists, Gödöllő, Hungary, 2022 
Posters 
1. The devil in the details: Eliminating false 
metagenomic classification with a novel 
algorithm 
Poster at the Hungarian Molecular Life 
Sciences 2023 Conference, Eger, Hungary, 
2023 
2. Understanding the effects of COVID-19 on 
the microbiome using bioinformatics and 
machine learning 
Poster at the 25th Spring Wind 
Conference,Pécs, Hungary, 2022 
3. Impacts of COVID-19 on the Microbiome: A 
Bioinformatics and Machine Learning Study 
Poster at the 21st European Conference on 
Computational Biology, Sitges, Spain, 2022 
NYlLATKOZAT
Nyilatkozom, hogy Takács Bertralan Vilmos társszeruőként vett részt a felelős
szerzőségemmel megjelent "Shotgun Analysis of Gut Microbiota with Body Composition and
Lipid Characteristics in Crohn's Disease" (B|OMEDICINES, 12 (9). ISSN 2227-9059 ea24\|
címú közleményben. A közlemény más PhD fokozatszerzésieljárásban társszerzőségként
nem volt felhasználva
Bertalan a közleményben az alábbi hozaájárulásokat végezte:
. Bioinformaükai analíás (shot-gun szekvenált mikrobiom minták összetéIelének
meghatározása}
. Statisáikaianalízis (Bray-Curtis távolság sámítás és főkoordinák-analízis)
. Adatvizualizáció
Anita Ph.D.


Development of a novel metagenomic classification method

Villy Lajos Péter

Hungarian

Kétkomponensű nanoszerkezetek szikra ablációs előállítása

Abdallah Hiba Faroug Muddather

Evaluating the in vitro anticancer potential of natural products and repurposed drugs in human cancer cell lines with different HPV status

Korir Evans Kiptoo

Understanding impact of Social Contact Patterns in Epidemiological Models

Karai Gábor

Topology preservation and skeletonization on the BCC and FCC grids

Paczona Viktor

Modern imaging methods in radiation therapy planning

Bacsosz Sztavrosz

A földrajzi tér kockázatcsökkentő hatásának vizsgálata a pénzügyi piacokon

University of Szeged

Not a member yet