=Paper=
{{Paper
|id=Vol-2491/abstract41
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-2491/abstract41.pdf
|volume=Vol-2491
|dblpUrl=https://dblp.org/rec/conf/bnaic/LibinVAGLN19
}}
==None==
<pdf width="1500px">https://ceur-ws.org/Vol-2491/abstract41.pdf</pdf>
<pre>
     Extended Abstract - Towards a phylogenetic
         measure to quantify HIV incidence
     Pieter Libin ∗1,2 , Nassim Versbraegen∗5,6 , Ana B. Abecasis2,3 ,
         Perpetua Gomes4 , Tom Lenaerts1,5 , and Ann Nowé1
1
  Artificial Intelligence Lab, Department of computer science, Vrije
                  Universiteit Brussel, Brussels, Belgium
 2
   Department of Microbiology and Immunology, Rega Institute for
    Medical Research, KU Leuven - University of Leuven, Leuven,
                                   Belgium
      3
        Global Health and Tropical Medicine, GHTM, Instituto de
Higiene e Medicina Tropical, IHMT, Universidade Nova de Lisboa,
                           UNL, Lisboa, Portugal
   4
     Laboratorio Biologia Molecular, LMCBM, SPC, HEM, Centro
                        Hospitalar Lisboa Ocidental
5
  Machine Learning group, Université Libre de Bruxelles, Boulevard
               du Triomphe CP212, 1050 Bruxelles, Belgium
         6
           Interuniversity Institute for Bioinformatics in Brussels,
                    ULB-VUB, 1050 Brussels, Belgium



    About 37 million people are currently infected with HIV and an estimated
35 million people have died due to the effects of AIDS (the eventual result of an
untreated HIV infection) since the beginning of the epidemic at the start of the
twentieth century. Global efforts have ensued to enhance the collection, dissem-
ination and accessibility of epidemiological data related to HIV epidemics. One
of the most burdensome aspects in curtailing the spread of HIV emerges from
infected individuals being unaware of their infection status. This stems from
the fact that a human host can be infected for many years before noticing any
symptoms. As a result, a significant fraction of the HIV infected population
remains undiagnosed, hampering effectiveness of interventions and assessment
of further developments of the epidemic. Consequently, methods that deliver a
well-founded estimate of the number of HIV infected individuals are paramount.
    ∗ Equal contribution.

    Copyright c 2019 for this paper by its authors. Use permitted under Creative Commons
License Attribution 4.0 International (CC BY 4.0).


                                           1
Such an estimate enables deduction of the number of undiagnosed infected in-
dividuals. State-of-the-art methods that aim to provide estimates of the size
of HIV epidemics generally consist of applying compartment models to routine
surveillance data to estimate the number of infected individuals (i.a., number
of new diagnoses over time and CD4+ cell counts).
    An abundance of clinical data is available in the context of HIV epidemics,
as upon diagnosis a number of tests are performed and the results thereof col-
lected. One of those tests comprises the assessment of the genotype of the virus
that infects the patient. To that extent, the genetic sequence of the virus is
determined. As a result, a vast amount of HIV sequences have been collected
over the last decades.
    The main benefit of developing a method to quantify a HIV epidemic that
relies on genetic data is to gain insight into the specific sub-populations that
contain a high rate of undiagnosed individuals, thus allowing for more effective
health policies, through diagnosis strategies that are directed towards these par-
ticular sub-populations.
We validate our research on the HIV-1 epidemic in Portugal. We therefore first
present inference of the epidemiological parameters of said epidemic by applying
approximate Bayesian computation (ABC). We apply ABC to fit a model that
incorporates the epidemiological parameters in question. We further show that
calibrating simulations to specific epidemics is essential, as the epidemiological
dynamic has an important impact on the shape of the phylogenetic tree. We
then construct a tree statistic that enables differentiation of epidemiological pa-
rameters based on phylogenetic trees and evaluate it on a set of epidemiological
simulations. We show that the presented tree statistic enables differentiation
of epidemiological parameters, while only relying on phylogenetic trees, thus
enabling the construction of new methods to ascertain the epidemiological state
of an HIV epidemic. By using genetic data to infer epidemic sizes, we expect
to enhance understanding of the portions of the infected population in which
diagnosis rates are low.




                                        2

</pre>