SCS colloquium: Incorporating detailed information on treatment history improves prediction of response to anti-HIV therapy

Title:Incorporating detailed information on treatment history improves prediction of response to anti-HIV therapy

Summary:

Infections with the human immunodeficiency virus type 1 (HIV-1)
are treated with combinations of drugs. HIV responds to the
treatment by developing resistance mutations.
For ensuring an effective treatment the genome of the viral target
proteins is sequenced
and inspected for resistance mutations.
For predicting response to a combination therapy, currently
available computer-based methods rely on the genotype of the virus
and the composition of the regimen as input. However, they do
not take full advantage of the knowledge about the order of and the response to
previously prescribed regimens. For exploiting such knowledge
the resulting high-dimensional feature space has to be explored efficiently.
Feature selection is essential for dealing with a problem of such
large dimensionality.
We present a classifier that employs L1 regularization (LASSO)
to maintain the sparseness of the solution.
In our algorithm, called sequence boosting, the sequence mining is
executed iteratively, and the feature space is extended progressively.
The resulting set of features is optimal with respect to
classification performance.
In computational experiments we show that the use of information on
treatment history
improves the prediction performance especially for patients with long
treatment records.
The medical relevance of the resulting discriminative treatment
patterns is discussed.