Protein Inhibotor resistance sites

Hypothesis:

Single site mutations in response to drug carry large fitness cost.

Compensatory mutations are needed for viable resistant strains.

Test:

Construct maximum entropy prevalence landscape (database www.hiv.lanl.gov)

• HIV-1 clade B Protease, removing insertions relative to HXB2
• Exclude all sequences after 1996, or with "protease inhibitor" in meta data [6701 from 757 patients]
• (Separate analysis with "drug naive" sequences)

Strong direct couplings indicate compensatory mutations of fitness close to wild type:

Eigen model of evolution with 2 sites:

Prune all direct couplings less than a threshold:

Standard resistance starts at site 82 followed by 54; the third strongest coupling above.

[PI Resistance sites, from M.E. Quiñones-Mateu1 and E.J. Arts, HIV Reviews 134 (2001)] [Rhee et al, NAR31, 298 (2003), Fig1]

Performance as a classifier:

• PPV:    Positive Prediction Value = Prob.[resistant is correct | predicted resistant ]
• NPV:    Negative Prediction Value = Prob.[predicted non-resistant is correct | predicted non-resistant ]
  • Larger PPV and NPV indicate more predictive power
• TPR:    True Positive Rate = Prob.[predicted resistant is indeed resistant]
• FPR:    False Positive Rate = Prob.[predicted resistant is non-resistant ]
  • At large values of the threshold, drug resistance sites are identified with high probability

[T.C. Butler, J. Barton, M.K., A.K. Chakraborty (2014) & SI] (Fig. SI5)

(9 out of top 40 pairs correspond to contacts in the dimer) (ala Martin Weigt)

Many other approaches to predicting drug resistance sites:

• Selection under treatment: L. Chen, A. Perlina, CJ Lee, J Viriology 78, 3722 (2004)
• Supervised learning:  CW Cao et al, Drug Discovery Today 10, 521 (2005)
• Structural modelling:   N Beerenwinkel et al, PNAS 99, 8271 (2002)
• ...

Next (EMF dynamics) Outline Adaptive Immunity Thymic Selection Quorum sensing Affinity Maturation Broadly neutralizing Abs EMF dynamics Drug Resistance Summary