Prediction of HIV viral tropism based on NGS data Nico Pfeifer Max - PowerPoint PPT Presentation

Prediction of HIV viral tropism based on NGS data Nico Pfeifer Max Planck Institute for Informatics

Cell entry Wu et al. Structures of the CXCR4 Chemokine GPCR with Small-Molecule and Cyclic Peptide Antagonists Science 19 November 2010: 330 (6007), 1066-1071.

V3 loop binds to coreceptor Wu et al. Structures of the CXCR4 Chemokine GPCR with Small-Molecule and Cyclic Peptide Antagonists Science 19 November 2010: 330 (6007), 1066-1071.

HIV tropism • Relevant coreceptors: CCR5 and CXCR4 • Viruses that can only use the CCR5 coreceptor: R5 • Viruses that can use the CXCR4 coreceptor: X4-capable

Entry inhibitors • Maraviroc – CCR5 antagonist – Approved for patient treatment • AMD-3100 – CXCR4 antagonist – Never approved for patient treatment

Want to know which patients benefit from taking maraviroc • Assays for tropism determination – Trofile – ESTA (enhanced sensitivity trofile assay) – Disadvantages: • Long turnaround • Require large sample volume • Genetic tests (V3 loop of gp120) – Sanger data – Next Generation Sequencing (NGS) data

Tools to predict tropism from genetic data • Sanger data – geno2pheno [coreceptor] [1] – WetCat [2] – WebPSSM [3] • NGS data – Variants of geno2pheno [coreceptor] and WebPSSM [4] 1. Lengauer T, Sander O, Sierra S, Thielen A, Kaiser R. Nat Biotechnol. , 2007 2. Pillai, S. et al . AIDS Res. Hum. Retroviruses 19, 145–149 3. Jensen, M.A. et al . J. Virol. 77, 13376–13388 4. Swenson, L. C. et al. J Infect Dis. (2011) 203 (2): 237-245

How do we represent the virus population inside a patient (V3 loop sequences)? CIRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTREGV-MGPG-AIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPG-AIYATGQIIGNIRQAHC CT---N--REGVHMGPG-AIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPG-AIYATGRIIGNIRQAHC CTR-NN-TREGVHMGPG-AIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPGGAIHATGQIIGNIRQAHC CTR-NN-TREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRANNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQARC CTRLNDNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTREGVHMGPGGAIYATRQIIGNIRQAHC CTRLNNNTREGVHMVPGGAIYATGQIIGNIRQAHC CTRLNN-TSEHISIGPGRAWVAARNIIGD-RKAHC CTRLNNNTRVGVHMGPGGAIYATGQIIGNIRQAHC CTRLNN-TSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTSE-ISIGPGRAWVAARNIIGDIRKAHC CTRLNNNT-EHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTSEHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTGEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARN-IGDIRKAHC CTRLNNNTNKHISIEPGRAWVAAREIKGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARNVIGDIRKAHC CTRLNNNTNKHISIGLGRAWVAAREIKGDIRKAHC CTRLNNNTSEHISIGPGRAWVVARNIIGDIRKAHC CTRLNNNTNKHISIGPGKAWVAAREIKGDIRKAHC CTRLNNNTSERISIGPGRAWVAARNVIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAR-IKRSIRKAHC CTRLNNNTSKHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARDIKGDIRKAHC CTRLNSNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREI-GDIRKAHC CTRLSNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRPNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRPNNNTRRSIHIGPGRAFYAG---IGDIRQAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHR CTRPNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDMRKAHC CTRPYANRKKSIHIGTG--FYTIKEIKGNVKQAYC CTRLNNNTNKHISIGPGRAWVAAREIKGGIRKAHC CTRPYANRKKSIHIGTGR-FYTIKEIKGNVKQAYC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC CTRPYANRRKSIHIGTG--FYTIKEIKGNVKQAYC CTRPYANRRKSIHIGTGR-FYTIKEIKGNVKQAYC CTRLNNNTNKHISIGPGRAWVAARNIIGGIRKAHC CTRPYANSRKSIHIGTG--FYTIKEIKGNVKQAYC CTRLNNNTNKHISIGPGRAWVAARNVIGDIRKAHC CTRVNNNTREGVHMGPG-AICATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRVNNNTREGVHMGPG-AIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRTWVAARQIIGDIRKAHC CTRVNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTNKHISLGPGRAWVAARNIIGDIRKAHC YTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC

Principal Component Analysis (PCA) • Represent axes of maximal variance (principal components)

Principal Component Analysis (PCA) • Represent axes of maximal variance (principal components) Principal component 1 (PC1)

PCA CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC

Next Generation Multi-Instance Learning Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRPYANRRKSIHIGTGRAFYTIKEIKGNVKQAYC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTREGVHMGPGRAIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Next Generation Multi-Instance Learning Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRPYANRRKSIHIGTGRAFYTIKEIKGNVKQAYC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTREGVHMGPGRAIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Next Generation Multi-Instance Learning Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRPYANRRKSIHIGTGRAFYTIKEIKGNVKQAYC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTREGVHMGPGRAIYATGQIIGNIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Improve predictions for last generation sequencing Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Improve predictions for last generation sequencing Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Improve predictions for last generation sequencing Patient 1 Patient 2 CTRLNNNTREGVHMGPGGAIYATGQIIGNIRQAHC CTRLNNNTSEHISIGPGRAWVAARNIIGDIRKAHC CTRLNNNTNKHISIGPGRAWVAAREIKGDIRKAHC CTRLNNNTREHISIGPGGAWVAAREIKGDIRKAHC CTRLNNNTNKHISIGPGRAWVAARQIIGDIRKAHC CTRLNNNTNKHISMGPGRAWVATGQIIGDIRQAHC CTRLNNNTNKHISIGPGRAWVAARNIIGDIRKAHC Support Vector Machine with normalized set kernel: 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑘 𝑙 𝑂𝑂𝑂 = � 𝑙 𝑡 𝑦 𝑗 , 𝑦 𝑗 𝑙 𝑡 ( 𝑦 𝑘 , 𝑦 𝑘 ) 𝑦 𝑗 ∈𝑌 𝑗 , 𝑦 𝑘 ∈𝑌 𝑘 Gärtner, T., Flach, P. A., Kowalczyk, A., Smola, A., J., Multi-Instance Kernels . International Conference on Machine Learning

Data • Maraviroc versus Optimized Therapy in Viremic Antiretroviral Treatment- Experienced Patients (MOTIVATE) + 1029 – 876 patients with NGS data of V3 loop • Also patients with X4-capable viruses (according to Trofile) – Treatment: maraviroc once-daily/twice-daily – Viral loads measured at various time points Swenson, L. C. et al. J Infect Dis. (2011) 203 (2): 237-245

Performance comparison • Predict class label: Treatment success • Compare measures in patient classes – Median log10 reduction in pVL after eight weeks • 5-fold nested cross-validation