1 / 29

Peircean Classification of Reasoning

Abductive Logic Programming in the clinical management of HIV/AIDS (and other domains) Oliver Ray, University of Bristol & Antonis Kakas, University of Cyprus FJ Symposium, September 2007, Aix-en-Provence.

alvis
Télécharger la présentation

Peircean Classification of Reasoning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Abductive Logic Programming in the clinical management of HIV/AIDS (and other domains)Oliver Ray, University of Bristol&Antonis Kakas, University of CyprusFJ Symposium, September 2007, Aix-en-Provence

  2. Abductive Logic Programming in the clinical management of HIV/AIDS (and other domains)Oliver Ray, University of Bristol&Antonis Kakas, University of CyprusFJ Symposium, September 2007, Aix-en-Provence

  3. Peircean Classification of Reasoning consequence: from prior knowledge to necessary implications Deduction (LP) Analytic Reasoning generalisation: from observed samples to wider populations: general rules Induction (ILP) Synthetic Reasoning explanation: from given effects to possible causes ground facts Abduction (ALP)

  4. Abductive Logic Programming • given • T - theory • G - goal • A - abducibles • IC - integrity constraints • find •  - explanation •  - answer • where • T  |*G - explanation • T  |*IC - integrity •   A - ground facts

  5. Worldwide Distribution of HIV Eastern Europe & Central Asia 1.5 million [1.0 – 2.3 million] Western & Central Europe 720 000 [550 000 – 950 000] North America 1.3 million [770 000 – 2.1 million] East Asia 680 000 [420 000 – 1.1 million] North Africa & Middle East 440 000 [250 000 – 720 000] Caribbean 330 000 [240 000 – 420 000] South & South-East Asia 7.6 million [5.1 – 11.7 million] Sub-Saharan Africa 24.5 million [21.6 – 27.4 million] Latin America 1.6 million [1.2 – 2.4 million] Oceania 78 000 [48 000 – 170 000] 40 million carriers [unaids, 2005]

  6. Structural and genetic makeup of HIV (2) host CD4 cell (3) viral structure (1) infected patient (4) viral genome

  7. HIV Disease Progression Immune Health (CD4 count) viral reproduction (Plasma Viral Load)

  8. Disrupting the HIV replication Cycle 3 2 (NRTI’s / NNRTI’s) 4 (PI’s) (FI’s) 1

  9. FDA-Approved Anti-Retrovirals (1987-2006) 3-4 drugs needed for Highly Active Anti-Retroviral Therapy (HAART)

  10. HIV Drug Resistance • Most drugs target the reverse transcriptase and protease enzymes • Copying errors in the viral genome case mutations in these enzymes • Some mutations confer resistance against (one or more) drugs • The patients therapy fails and a salvage treatment must be found (1) Resistance mutations in Rev- erse Transcriptase (e.g. K103N) (2) Resistance mutations in Protease (e.g. V82A)

  11. Laboratory HIV Resistance Tests 1. Genotypic tests (identify resistance conferring mutations in viral genes ) 2. Phenotypic tests (measure n-fold resistance in lab-cultured assays)

  12. HIV Genotypic Interpretation Rules Resistance rules are published by leading AIDS research institutes including ANRS (Paris), REGA (Leuven), and HivDB (Stanford)

  13. Stanford Algorithm Specification Interface Raw ANRS / REGA / HIVDB rules (AZT) XML Stanford Algorithm Specification Interface

  14. Limitations of Resistance Testing • Cannot detect minority and archived Strains • are insensitive to strains comprising less than 10% of a patients viral population (even tough these strains can persist undetected for years and harbour drug resistant mutations) • Are expensive and require hi-tech equipment • each test costs 250$-750$ and requires access to sophisticated laboratory machinery (to which most HIV infected individual do not have access) • Hence careful interpretation is needed and a way of predicting resistance from clinical data in the absence of such tests is highly desirable.

  15. In-Silico Sequencing System (iS3) HIV resistance model 3 6 1 predicted drug resistance patient data 5 4 in-silico sequencing abductive explanations 2 7 • Use genotypic rules abductively to infer mutations from clinical data • Use statistical methods to extract predictions from possible answers predicted and known mutations geno rules

  16. Summary of effective and ineffective treatments as determined by doctor • Automatically extracted and processed from clinical database 1 patient data unknown (p056, 1, [AZT,3CT,IDV ]). ineffective (p056, 2, [AZT,3CT, IDV]). ineffective (p056, 3, [D4T,DDI,SQV,RTV]). effective (p056, 4, [EFV,SQV,RTV]). ineffective (p056, 5, [IDV,EFV,RTV]). effective (p056, 6, [LPV,EFV,DDI ]). genotype (p056, 7,[184V, 69D, 70R, 41L, 215Y, 30N]). effective (p056, 8, [3CT,RTV,DDI,ATV]).

  17. resistant(AZT) :- present (1, [215YF,151M,69i]). resistant(AZT) :- present (3, [41L,67N,70R,210W,215ACDEGHILNSV,219QE]). resistant(DDI) :- present (2, [41L, 69D, 74V, 215FY, 219EQ]), present (2, [~184IV, ~70R)]). • logical encoding of ANRS AC11 genotypic HIV drug resistance interpretation rules. • Automatically downloaded and extracted from the Stanford HIV Database • Note group mutations e.g. 219EQ • Note antagonistic mutations e.g. ~70R 2 geno rules

  18. HIV resistance model • Commonsense principles and working assumptions about drug resistance 3 ineffective(Patient, Time, Drugs) :- in(Drug, Drugs), resistant(Patient, Time,Drug) effective(Patient, Time, Drugs) :- not ineffective(Patient, Time, Drugs) mutation(Patient, Time1, X) :- mutation(Patient, Time2, X), Time1 >= Time2

  19. Use rules abductively (in reverse) to explain patient data in terms of mutations they may be carrying by using their clinical history • Process time-points incrementally, storing the minimal explanations from the previous time-point for future use • Minimality: don’t hypothesise more mutations than necessary to explain the data. Assumes treatment failures are detected early. ineffective(p056,2,[ AZT,3TC,IDV ]) [mutation(p056,2,215YF)] [mutation(p056,2,151M)] [mutation(p056,2,69i)] ineffective(P,T,Ds) :- in(D,Ds), resistant(P,T,D) 4 is3 minimal abductive explanations resistant(AZT):-present (1,[215YF,151M,69i]).

  20. Extract statistical information from the (thousands) of explanations • Warn doctor if he prescribe a drug with high predicted resistance score each drug according to how many explanations imply its resistance (by using the interpretation rule forwards) 6 [mutation(p056,2,215YF)] [mutation(p056,2,151M)] [mutation(p056,2,69i)] predicted drug resistance score each mutation according to how many explanations it appears in (giving a higher weight to shorter explanations) 7 predicted and known mutations

  21. HIV Resistance Analyser general data: patient ID and visit current meds: list updated from HIV-DB genotype results if available assessment of the current meds as determined by an expert

  22. Evaluation 1: Predicted Mutations 100 80 60 40 20 0 20 40 60 80 100 0 % actual mutations detected by genotype Useful Clinical cutoff ? top 1/3 of predicted mutations contain 2/3 of observed mutations % top ranked mutations predicted by system n.b. here, the mutation rankings are post-processed to account for selection pressure resulting from the drugs taken at the time of the genotype

  23. Evaluation 2: Predicted Resistance • Run system up to (but not including) the last time when the treatment changed and a definite outcome was observed • Compare the system’s predictions with the known outcome

  24. Summary HIV resistance model 3 6 1 predicted drug resistance patient data 5 4 in-silico sequencing abductive explanations 2 7 • Practical application of ALP • Method for multiple solutions • potential for clinical use predicted and known mutations geno rules

  25. Future Work more clinical & genotype test data is needed more testing on clinical data mutation pathways extended HIV resistance model 3 6 1 predicted drug resistance mutation decay meta-level reasoning improved statistics patient data 5 4 more testing on clinical data compare quality of other genotypic rules in-silico sequencing abductive explanations 2 7 predicted and known mutations geno rules

  26. Related Work: Gene regulation networks (Papatheodorou, Sergot, Kakas - LPNMR’05) TB/Yeast databases 3 4 3 abductive explanations ALP micro-array data mutant strains and environmental shock gene interaction networks

  27. Related Work: Inhibition in Metabolic Networks (Tamadonni-Nezhad, Chaleil, Muggleton, Kakas - ML) KEGG 3 4 3 abductive explanations ALP/ILP enzyme inhibition hypotheses nmr data from rodent urine

  28. Related Work: Robot Scientist(King et al. - Nature)

  29. END

More Related