imagine (Institute of molecular genetics and genetic engineering)
Not a member yet
3088 research outputs found
Sort by
Future directions in network biology
No full textNetwork biology, an interdisciplinary field bridging computational and biological sciences, has
revolutionized understanding of cellular functions and diseases. The field, which has existed
for two decades, has witnessed rapid evolution, accompanied by emerging challenges. These
challenges stem from various factors, notably the growing complexity and volume of data
together with the increased diversity of data types describing different scales of biological
organization.
This talk will discuss some of key research areas in network biology and highlight recent
breakthroughs in these areas; offer a perspective on the future directions of network biology;
and touch on scientific communities, educational initiatives, and the importance of fostering
diversity within the field.
Two specific research directions will be discussed. The first is on our network-of-networks
analyses of multi-scale biological systems with application to protein function prediction. The
function of a protein is determined by the protein’s 3D structure, which also affects which
other proteins the protein interacts with. Because of this, and because nodes in a proteinprotein
interaction (PPI) network can be represented as protein structure networks (PSNs),
we modeled the integrated PPI and PSN data as a network-of-networks. We found that the
multi-scale network-of-network analysis often resulted in more accurate protein function
prediction than traditional single-scale analysis of PPI data alone or PSN data alone.
Second, the talk will discuss our network-based analyses of protein folding. We had proposed
several approaches for modeling protein 3D structures as PSNs. Static PSNs model the whole,
final 3D structure of a protein. Because the folding of a protein is a dynamic process, where
some parts (3D sub-structures) of a protein fold before others, most recently, we modeled a
protein as a dynamic PSN that captures these sub-structures. We evaluated our PSN models
in the task of protein structural classification. We found that our PSN models outperformed
state-of-the-art approaches for the same task, with dynamic PSNs being superior to static
PSNs.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
“Pan-viral” disease mechanisms unveil the sweet spot for therapeutic intervention
No full textViral infections continue to cause pandemics. Antiviral drugs are based on two strategies:
targeting the viral proteins or the host proteins. The main limitations of antiviral drugs
targeting viral proteins are their high ratio of drug resistance and their specificity to one virus.
A rapid and effective way to overcome these limitations is by re-purposing existing drugs that
target the host biological mechanisms used by the virus. Moreover, when these biological
mechanisms are shared across different viruses (i.e., “pan-viral” disease mechanisms), the
re-purposed drugs are good candidates for broad-spectrum viral treatments.
In our previous work, we investigated how the viral interactors (VIs) of SARS-CoV-2 and
differentially expressed genes (DEGs) after COVID-19 infection are connected in the host
interactome. We uncover that, in the host interactome, VIs and DEGs, while mostly disjoint,
are indirectly connected by their neighbours (we termed them “common neighbours” (CN)
genes). Furthermore, we found that the CN genes are key to COVID-19 mechanisms and
promising targets for drug re-purposing.
Here, we expand our results to 13 viruses, uncovering “pan-viral” genes and re-purposing
drugs for broad-spectrum viral treatments. First, we identify CN genes for five well-studied
viruses and approximate their identification for viral infections without DEG data, uncovering
disease genes for 13 viruses (8 without DEG data). Then, we find that the CNs are shared
across viruses with significant enrichment in viral and immune system-related processes,
showing our methodology’s capability to uncover “pan-viral” disease mechanisms. Finally,
we predict new drug-target interactions to identify drugs targeting our “pan-viral” genes
by using a two-step machine-learning model. The CNs allowed us to re-purpose drugs for
treating viral infections by disrupting “pan-viral” disease mechanisms, paving the way for
broad-spectrum drug re-purposing. Moreover, the CN genes can enable insight into other
infectious diseases.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
Just feed it to the network - what can go wrong? Towards AI-supported early cancer detection from cytological image data
No full textCost-effective procedures for sample collection, as well as for their subsequent analysis,
can enable large scale screening programs for early detection of oral cancer, leading to
significantly improved patient survival. Brush samples taken from patients’ oral cavities
provide a solution to the first requirement (cost-effective sample collection). AI-supported
analysis of the whole slide images of Papanicolaou stained cytological samples has the
potential to offer a solution to the second (cost-effective analysis).
Can we simply train a deep neural network to classify the acquired whole slide images
into two classes, differentiating samples from patients with oral cancer and samples from
healthy patients?
It appears that there are many challenges to address along that road. In this talk, I will
share our experience in developing methods for an AI system that can provide reliable
support for a cytologist and enable interpretable, fast and cost-effective early detection
of oral cancer in digital pathology.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
Feature Selection For Multi-Source SCT Data
No full textBioinformatics experiments often produce large data sets, with lots of samples and many
different attributes, which provide high dimensionality of the data. Even if some of the
dimensions have little significance for specific data analysis, they can prove useful in
complex data processing. Advanced data mining techniques and AI algorithms typically
welcome high dimensionality of data. However, if there are too many dimensions, we can
run into the curse of dimensionality, because an abundance of dimensions can introduce
additional complexity and cost to data handling and processing, as well as overfitting the
model to less important dimensions. In order to make data processing more efficient and
improve the quality of created models, we often need dimensionality reduction.
Single cell transcriptomics (SCT) is one of the most important sequencing technologies
today. It enables simultaneous measurement of the activity of thousand of genes in
individual cells, resulting in RNA profiles of the cells. Such profiles allow researchers to
analyze the physiological activity of cells in different circumstances, including different
biochemical conditions but also different health conditions of the subjects. By processing
large numbers of cells, SCT provides a lot of sample data. Each detectable RNA represents
one dimension of data, which ultimately gives us thousands of dimensions. Moreover,
initial cells conditions, complex cell preparation techniques, and RNA measurement
methods can vary significantly, resulting in significant differences in data coming from
different sources.
Here we discuss the feature selection problem on the example of more than 120,000
instances of peripheral blood mononuclear cell (PBMC) SCT data from four different
sources, with the detection of 30,698 genes (dimensions). We pay special attention to
the imbalanced nature of the data and consider feature selection methods that allow for
an unbiased set of significant features to be obtained as a result. We show that statistical
correlation-based feature selection, with some support from mutual information-based
techniques, can result in a reasonably complex method for high-quality feature set
selection.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 202
Feature Selection For Multi-Source SCT Data
No full textBioinformatics experiments often produce large data sets, with lots of samples and many
different attributes, which provide high dimensionality of the data. Even if some of the
dimensions have little significance for specific data analysis, they can prove useful in
complex data processing. Advanced data mining techniques and AI algorithms typically
welcome high dimensionality of data. However, if there are too many dimensions, we can
run into the curse of dimensionality, because an abundance of dimensions can introduce
additional complexity and cost to data handling and processing, as well as overfitting the
model to less important dimensions. In order to make data processing more efficient and
improve the quality of created models, we often need dimensionality reduction.
Single cell transcriptomics (SCT) is one of the most important sequencing technologies
today. It enables simultaneous measurement of the activity of thousand of genes in
individual cells, resulting in RNA profiles of the cells. Such profiles allow researchers to
analyze the physiological activity of cells in different circumstances, including different
biochemical conditions but also different health conditions of the subjects. By processing
large numbers of cells, SCT provides a lot of sample data. Each detectable RNA represents
one dimension of data, which ultimately gives us thousands of dimensions. Moreover,
initial cells conditions, complex cell preparation techniques, and RNA measurement
methods can vary significantly, resulting in significant differences in data coming from
different sources.
Here we discuss the feature selection problem on the example of more than 120,000
instances of peripheral blood mononuclear cell (PBMC) SCT data from four different
sources, with the detection of 30,698 genes (dimensions). We pay special attention to
the imbalanced nature of the data and consider feature selection methods that allow for
an unbiased set of significant features to be obtained as a result. We show that statistical
correlation-based feature selection, with some support from mutual information-based
techniques, can result in a reasonably complex method for high-quality feature set
selection.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
Challenges in metagenomic annotation of antibiotic resistome
No full textThe antibiotic resistance genes (ARGs) in both host-associated and environmental microbiomes –
antibiotic resistome – play an important role in the spread of antibiotic resistance. Metagenomics
enables high-throughput exploration of microbiomes. For ARGs annotation within the
metagenomes, several tools are in use, however, most of them are developed for genomics
studies and their databases may pose certain biases.
We compared outputs from different ARGs annotation approaches: we ran >13 000 high-quality
metagenomes from 14 habitats (Coelho et al., 2022) in three input modes (read/assembly/
protein) through six ARGs annotation pipelines using RGI, ABRicate, ResFinder, AMRFinderPlus,
DeepARG and ARGs-OAP. To facilitate comparison of outputs with different gene names, we
performed ARO (CARD’s Antibiotic Resistance Ontology) normalization using our tool https://
github.com/BigDataBiology/argNorm.
DeepARG and RGI provide higher coverage (with potential novel ARGs and/or false positives)
while ABRicate’s different databases have lower coverage but more well validated ARGs.
The annotations did not differ only in hits number but also in information provided: e.g. using
ABRicate, more sulfonamide ARGs were identified using ResFinderFG version 2.0 as the database
(Gschwind et al., 2023), while using ResFinder resulted in more macrolide ARGs. Thus, choice of
annotation tool and database should be driven by research questions and ARGs targets.
Coelho, L.P., Alves, R., del Río, Á.R. et al. Nature (2022). doi: 10.1038/s41586-021-04233-4
Gschwind, R., Ugarcina Perovic, S., Weiss, M. et al. Nucleic Acids Research (2023). doi: https://doi.
org/10.1093/nar/gkad384Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
The landscape of point mutations leading to pregnancy loss
No full textMiscarriage is the most frequent complication of a pregnancy which is devastating for
affected families and poses a significant challenge for the health care system. Genetic
factors are known to play an important role in the etiology of pregnancy loss; however,
despite advances in diagnostics, the causes remain unexplained in more than 30% of
cases.
In this work, we aggregated the results of the decade-long studies into the genetic risk
factors of sporadic spontaneous abortion and recurrent pregnancy loss (RPL) in euploid
pregnancies, focusing on the spectrum of causal point mutations in the fetal genome.
A total of 270 genetic variants in 196 unique genes were identified across 31 studies,
with the majority of these variants associated with non-recurrent miscarriage. We next
aimed to reveal the common properties of the identified genes involved in pregnancy loss.
Such an analysis showed that they correspond to broadly expressed, highly evolutionary
conserved genes involved in crucial cell differentiation and developmental processes and
related signaling pathways. Given these features of known genes, we made an effort to
construct a list of 186 candidate genes, variants in which may be expected to contribute
to pregnancy loss. We subsequently expanded this work and compiled a database which
includes all short genetic variants reported as associated with pregnancy loss. To date,
the database includes 479 genetic variants observed in 374 cases, including both euploid
fetuses and couples experienced recurrent pregnancy loss.
Taken together, our work summarizes the knowledge about specific genes that contribute
to pregnancy loss, and could be of importance in designing a diagnostic sequencing panel
for patients and prediction of pregnancy loss risk in couples.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
Bioinformatic workflow to analyse Multiome ATAC + Gene Expression data
No full textMulti systemic inflammatory syndrome in children (MIS-C) is a rare condition associated
with SARS-CoV-2. To improve the understanding of underlying regulatory networks
and potentially explain the mechanism behind MIS-C disease onset, we conducted a
simultaneous profiling of transcriptome and epigenome from the same cell using the 10X
Genomics Chromium Single Cell Multiome ATAC + Gene Expression protocol. Our study
design consisted of 10 patients sampled at two time points: at the time of MIS-C disease
flare before treatment was applied, and at the time of disease remission.
In order to facilitate an efficient way of processing the wealth of data generated by
the employed protocol, we developed a bioinformatic workflow that enables users
to systematically extract information crucial for understanding cellular function and
regulation. The workflow is comprised of primary analysis performed using 10X Genomics
Cell Ranger ARC software, and advanced analysis conducted utilizing a curated collection
of various R packages. Through the proposed analyses, researchers can effectively
identify and correct technical aberrations and batch effects in the data, unveil distinct
cell types, detect differential gene expression between flare and remission states,
reveal heterogeneity within cell populations, pinpoint enriched biological pathways and
functions, and elucidate regulatory elements controlling gene activity.
The results of this project are expected to enlighten the underlying pathophysiology
of MIS-C flare and support clinical decision on more targeted treatment. The identified
disrupted networks during MIS-C flare could lead the way to establish an early
diagnosis and improve long-term outcome, including prevention of myocardial and
neuropsychological impairment.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
A systematic characterization of genotype-to-phenotype relationships in mouse and human
No full textReconstruction of the genotype-to-phenotype relationship network is one of the major
goals of modern genomics. Development of ontologies for accurate and formal phenotype
description enables a systematic comparison of these relationships between species. In
this work, we employed data on genotype-to-phenotype associations from the Human
Phenotype Ontology (HPO) and Mouse Genome Informatics (MGI), as well as genomewide
associations from the UK Biobank (UKB) cohort to investigate the similarities and
dissimilarities in the architecture of genotype-to-phenotype associations in mice and
humans.
Comparison of the sets of upper-level Mammalian Phenotype Ontology (MP) terms for
16,985 orthologous gene pairs showed that only 24.6% (4,184) of all gene pairs were
annotated with phenotype terms in both species. Even more surprisingly, only a handful
of 15 gene pairs were annotated with exactly identical terms in human and mouse. Of
the remaining ones, as many as 385 genes were associated with non-overlapping sets of
phenotypes in humans and mice, and as many as 3,784 genes had both concordant and
discordant gene-trait associations.
Despite such large differences in genotype-to-phenotype relationships for individual genes,
we found that the overall architecture of the genotype-to-phenotype network was similar
for the two species. For instance, the pleiotropic effects of orthologous genes were well
correlated (Spearman’s p = 0.28) when using upper-level phenotype term count as a measure
of pleiotropy. In both species, higher degree of pleiotropy also correlated with various genelevel
properties, such as greater evolutionary constraint of a gene and its broader expression
across tissues. Similar trends were observed when using UKB genome-wide associations to
estimate the degree of pleiotropy for human genes.
Taken together, our observations highlight important discrepancies between phenotype
description in humans and mice. At the same time, our analysis suggests that the
organization of genotype-to-phenotype networks is largely similar for highly related species.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024
Analysis of structural features of DNA in tRNA genes
No full textRNA polymerase III (Pol III) transcribes tRNA genes using type 2 promoters, namely
intragenic boxes A and B. In addition, 5’-flanking regions of tRNA genes of plants and
S. pombe contain octanucleotides similar to the well-known TATA-box in the promoters
of mRNA genes. The TATA box has not been found in other eukaryotes, although TBP is
the component of the transcription factor TFIIIB. Archaea use the orthologue. The goal
of this work was to determine the position of possible TBP binding while tRNA genes
transcription in various eukaryotes as well as the comparison of structural properties of
nucleotide sequences of tRNA genes and their up-stream regions in eukaryotes, archaea
and bacteria. We have analyzed representative sets of tRNA gene sequences from 11
organisms from the Genomic tRNA Database (GtRNAdb) (http://gtrnadb.ucsc.edu). The
nature of the upstream DNA sequences has been discussed earlier. It was proposed to
consider these areas as a fine-tuning control of the transcriptional activity and starting
specificity of a given tRNA gene. The results of experimental work already existing by
that time led to the conclusion that transcription initiation of eucaryotic tRNA genes thus
appears to depend critically on DNA conformation, both within and around the genes. The
characteristics of textual and spatial structure in the vicinity of the start of tRNA genes
were detected. Textual characteristics of 60 bp sequences aligned to the ends of the genes
were also analyzed and the position of B-box was defined. It was found that belonging
to different domains has little effect on the consensuses of A and B boxes. They differ
only at the degree of conservatism of some positions. The mechanical properties of the
upstream regions of all eukaryotes allow TBP to bind not in one, but in several positions,
but with different affinities. These properties are less pronounced in archaea, and absent
in bacteria.Book of abstracts: 5th Belgrade Bioinformatics Conference, Serbia, Belgrade,17-20 june 2024