Prediction of co-receptor usage of HIV-1 from genotype

Abstract

Human Immunodeficiency Virus 1 uses for entry into host cells a receptor (CD4) and one of two co-receptors (CCR5 or CXCR4). Recently, a new class of antiretroviral drugs has entered clinical practice that specifically bind to the co-receptor CCR5, and thus inhibit virus entry. Accurate prediction of the co-receptor used by the virus in the patient is important as it allows for personalized selection of effective drugs and prognosis of disease progression. We have investigated whether it is possible to predict co-receptor usage accurately by analyzing the amino acid sequence of the main determinant of co-receptor usage, i.e., the third variable loop V3 of the gp120 protein. We developed a two-level machine learning approach that in the first level considers two different properties important for protein-protein binding derived from structural models of V3 and V3 sequences. The second level combines the two predictions of the first level. The two-level method predicts usage of CXCR4 co-receptor for new V3 sequences within seconds, with an area under the ROC curve of 0.937+/-0.004. Moreover, it is relatively robust against insertions and deletions, which frequently occur in V3. The approach could help clinicians to find optimal personalized treatments, and it offers new insights into the molecular basis of co-receptor usage. For instance, it quantifies the importance for co-receptor usage of a pocket that probably is responsible for binding sulfated tyrosine.

Figure 1. Receiver Operating Characteristic (ROC) curves of the two-level random-forest classification approach.

Solid curves: averaged over ten-fold leave-one-patient-out cross-validation with random forests trained on interpolated Kyte-Doolittle hydrophobicity along normalized sequences (green), on electrostatics hull (red), and on probability outputs of the two previous random forests, i.e. second-level classification (blue); error-bars mark 95% confidence. Dashed curve: averages over ten out-of-bag predictions of second-level random forests on the full training set of sequences, disregarding that several sequences may originate from same patient.

Figure 2. 5% most important positions on electrostatics hull for tropism classification by electrostatics based random forest.

The backbone of the template V3 conformation is shown as tube with C_α atoms marked by small beads and some residues numbered for orientation, starting with the N-terminal Cys as residue 1. Points are colored according to the mean electrostatic potential (unit ) in the respective tropism class (red, ; light red, ; white, ; light blue, ; blue, ).

Figure 3. Importance of positions of normalized V3 sequence in random forest classification with Kyte-Doolittle descriptor .

The higher the peak at the respective position, the more important this position for correct classification of sequences with respect to co-receptor tropism. The most important region is around normalized sequence position 12, in agreement with the 11/25 rule. The second most important region around position 8 could be involved in binding of sulfated tyrosine on CCR5 . Along the top axis, reference sequence HXB2 before normalization is given for orientation.

Figure 4. X4 class probabilities for sequences as predicted by the two first-level random forests.

Vertical and horizontal axis give probabilities from electrostatics and hydrophobicity based random forests, respectively. These data points are the input for the second-level learning. Note that the sets of R5-tropic sequences (circles) and X4/R5X4-tropic (crosses) can be separated quite well in the plane spanned by the two descriptors.