Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Horizontal
gene
transfer
Background: Possible solutions to AMR problem
1. Antimicrobial compounds with new or improved modes of action
a. Improvement of resolving coding and noncoding regions and analysis of regulatory regions
Bacterial
membrane Conservation
of protein
(gene)
defines
functional
groups
CATAACAGGGGAA...
ATTGATTGAAAATA...
Functional AATATATCGCCAGC
groups
(color AGCACATGAACAAG...
TTTCGGAATGTGAT
coded)
CAATTTAAAAATTT...
ATTGACTTAGGCG...
GGCAGATACTTTAA...
Results: Analysis of DNA structural properties
1. Parametric and ML based models including
a. Thermodinamically induced duplex destabilization
b. Properties related to DNA-protein interaction
c. 64 such models
dsDNA
Results: Analysis of DNA structural properties
Permutations of 5 Analysis of
nucleotides: Prediction of 64 principal
AAAAA
AAAAC
structural properties components
AAAAG
...
and clustering
Properties of one
DNA DNA clusters
permutation
sequence structure
2 bits (4 clusters)
3 .
4 (16 clusters) PC3
2 bits
information 5 .
6 .
7 .
8 (256 clusters) PC2
PC1
Results: Analysis of DNA structural properties
Dataset of regulatory regions
CATAACAGGGGAA...
ATTGATTGAAAATA...
Functional AATATATCGCCAGC Vectors of
groups values of
(color AGCACATGAACAAG... structural
coded) TTTCGGAATGTGAT variables
CAATTTAAAAATTT...
ATTGACTTAGGCG...
GGCAGATACTTTAA...
Results: Statistical analysis of DNA representations
1. Variance ~ distance between elements
a. structures Euclidean distance
b. sequences p-distance = 1 - sum(identity)
nbins = 100
Frequency
FBootstra
p
Anderson 2001
Results: Statistical analysis of DNA representations
Structure
Sequence
Conservation
of data in
groups (F)
Amount Size of
of bits Size of region
region
Results: Prediction of transfer properties
- Models built using Neural Networks and
- 64 experimental regions, tested with 140 new regions:
accuracy = 0.975 0.004
- 10-fold cross validation (90% training and 10% testing examples)
with all 204 regions: accuracy = 0.945 0.008
Frequency
Pseudocode: nbins = 100
for i = 1 : size(query)
{ for j = 1 : length(target)
Distance d
{ d(i,j) = dist(query,target)
} % dist is Euclidean distance
}
find min(d(:,:) < treshold)
% treshold is set by statistical significance
dBoostrap
Position in target
Using this method we find new regulatory transfer regions with p < 10-6 in over 60% of targets.
Conclusions and further work
- New structural representation (code) for DNA
- Significant improvement with functional regulatory regions
- Webtools at www.dnatools.eu
- prediction of transfer properties and hosts
Thank you for listening
Acknowledgements
Dr. Ale Lapanje
Dr. Toma Rijavec
Dr. Tatjana Zrimec
Dr. Ale Beli
and others