Comparative Drug Ranking for Clinical Decision Support in HIV Treatment Emiliano Mancini University of Amsterdam 1
Prevalence of HIV 2
Global Overview HIV Infection People living with HIV: 34 million (2010 & 2012) New HIV infections: 2.7 million (2010) - 2.5 (2012) million AIDS- related deaths: 1.8 million (2010) 2.3 (2012) million People receiving Antiretroviral Therapy 6.5 million (2010) 15 million (2015) HIV Treatment Source: UNAIDS 3
P.M.A. Sloot et al., HIV Decision Support: From Molecule to 2 patents and in use in 12 EU Man, Phil. Trans. R. Soc. A, vol. 367, nr 1898 pp. 2691-2703. Hospitals HIV CCR5 CXCR4 Agent-Based Entry Simulation CD4+ Target Cell CA Based Immune Response Phenotype Complex Networks Epidemics Molecular Dynamics Binding Affinity Protein Structure & Binding Affinity Protease and RT mutations DRUG RANKING DECISION SUPPORT Clinical Parameters: -weight - opportunistic infections and tumors -survival Text Mining è Drugranking ç 1 st order logic 4
Complex Interplay in HIV 5
Simulation Results Validation of Dynamics Predictive models CD4+ cells /mm 3 Continuous Therapy CD4+ cells /mm 3 Optimal Therapy CD4 + T 1 0.8 0.6 0.4 CD4+ cells /mm 3 Provirus virion /mm 3 8 weeks on / 4 weeks off CD4+ cells /mm 3 Provirus virion /mm 3 No therapy 0.2 Provirus virion /mm 3 Provirus virion /mm 3 2 4 6 8 10 12 years P.M.A. Sloot et al., 2003, 2005, 2010 E.Mancini et al., 2012 6
HIV Life Cycle 7
HIV RNA virus 10 9 new viruses produced every day RT makes errors during each transcription Due to high error rate, multiple mutations 8
Accumulation of high level resistance Number of Mutations Accumulated 2 mutations 4- fold resistance 3 mutations 10- fold resistance (!) 9
Drug Resistance 250,000 HIV individuals in the USA and Europe with drug resistant viruses. 10% of new infections in USA/Europe occurs with viruses that have at least one drug resistant mutation. Small proportion of patients die because we have no drugs to inhibit their viruses 10
Choosing a HIV therapy Clinicians take informed decisions Yearly Guidelines + Knowledge + Experience that are driven by intuition Clinicians have on average ~5 minutes per patient! 11
How to choose the right therapy? Complex task Short time HIV Genotype and mutations Drugs side effects Drug resistances Drugs interactions Cell counts Errors Consequences Treatment failure à HIV mutations selected More mutations à more treatment failures How to minimize those errors? 12
Drug Resistance Databases HIVdb (Stanford) ANRS (France) REGA (Leuven) Input Output Virus Genotype Level of drug resistance Issue Discordances between them What is the next step? 13
Decision Support System Comparative Drug Ranking System Drug Ranking Component Display Resistances for drugs tailored to individual patients Identifies Discordances between rulesets Literature Mining Component Provides tools to resolve Discordances 14
Comparative Drug Ranking System Drug Ranking Component First order logic to translate and compare rules from different drug ranking rulesets Consistency checks inter and intra rulesets Highlight Discordances 15
Comparative Drug Ranking System Input a viral genome or a set of mutations 16
Drug Ranking Component Alert icon to highlight discordances Display specific rules triggered by input muta,ons Intui,ve visualiza,on of the score Legend of the drug resistance ranking Literature Mining Tool 17
Comparative Drug Ranking System Literature Mining Component Search in PubMed database for relevant papers Extract relations between drugs and specific mutations in analysis Rank papers by relevance to the mutations of the individual patient in analysis 18
Literature Mining Component A novel method 1 to extract and combine causal relations of mutations on HIV drug resistance. Based on Natural Language Processing which produces grammatical relations and applies a set of rules to these relations. Automatically mine > 30.000 PubMed documents and obtain 9,427 causal relations and 3034 sentences with relations Combine the extracted relations using logistic regression and generate resistance values for each <drug, mutation> pair Bui; Ó Nualláin; Boucher, Sloot: Extracting causal relations on HIV drug resistance from literature, BMC Bioinformatics 2010. Bui; S. Katrenko, P.M.A. Sloot: A hybrid approach to extract protein- protein interactions, Bioinformatics 2011. 19
Literature Mining Component 20
Conclusions - 1 Emergence of Drug Resistant mutations Suboptimal treatment à drug resistant mutations Treatment failure HIV mutations selected More treatment failures More mutations Prevent accumulation of high level Drug Resistances 21
Conclusions - 2 Choose the appropriate treatment Display level of resistance for each drug Identify discordances between databases Provide evidence to resolve discordances Clinical Decision Support Deployed on VPH- Share platform 22
Acknowledgements Peter Sloot Fajran Rusadi Piotr Nowakowsky VPH- Share Funded by: European Commission FP7 Contract no: 269978 Thank you! Questions? 23