Machine Learning in Drug DesignCollaboratorsOutlineDrugs Typically Are…Example of BindingSo To Design a Drug:Molecule Binds Target But May:And Every Body is Different:Slide 9Places to use Machine LearningSlide 11Healthy vs. DiseaseIf We Could Sequence DNA Quickly and Cheaply, We Could:Problem: Can’t Sequence QuicklySlide 15Example of SNP DataProblem: SNPs are not GenesProblem: Even SNPs are CostlyWhy Machine Learning?Slide 20Decision Trees in One PicturePowerPoint PresentationNaïve Bayes in One PictureVoting ApproachTask: Predict Early Onset Disease From SNP DataResultsLessonsSlide 28Slide 29Typical Practice when Target Structure is UnknownAn Example of Structure LearningInductive Logic ProgrammingThe Logical Representation of a PharmacophoreBackground Knowledge IBackground knowledge IICentral Idea: Generalize by searching a latticeConformational modelPharmacophore descriptionExample 1: Dopamine agonistsPharmacophore identifiedExample II: ACE inhibitorsExperiment 1ACE pharmacophorePharmacophore discoveredExperiment 2Slide 46Example III: Thermolysin inhibitorsKey binding site interactionsInteractions made by inhibitorsPharmacophore IdentificationThermolysin ResultsThermolysin resultsExample IV: Antibacterial peptidesPharmacophore IdentifiedSlide 55Ongoing ILP developments (pharmacophores)Ongoing developments (Other)Machine Learning in Drug DesignMachine Learning in Drug DesignDavid PageDept. of Biostatistics and Medical Informatics and Dept. of Computer SciencesCollaboratorsCollaboratorsMichael WaddellPaul FinnAshwin SrinivasanJohn ShaughnessyBart BarlogieFrank ZhanStephen MuggletonArno SpatolaSean McIlwainBrian KayOutlineOutlineOverview of Drug DesignHow Machine Learning Fits Into the ProcessTarget Search: Single Nucleotide Polymorphisms (SNPs)Machine Learning from Feature VectorsDecision TreesSupport Vector MachinesVoting/EnsemblesPredicting Molecular Activity: Learning from StructureDrugs Typically Are…Drugs Typically Are…Small organic molecules that…Modulate disease by binding to some target protein…At a location that alters the protein’s behavior (e.g., antagonist or agonist).Target protein might be human (e.g., ACE for blood pressure) or belong to invading organism (e.g., surface protein of a bacterium).Example of BindingExample of BindingSo To Design a Drug:So To Design a Drug:Identify TargetProteinDetermineTarget SiteStructureSynthesize aMolecule thatWill BindKnowledge of proteome/genomeRelevant biochemical pathwaysCrystallography, NMRDifficult if Membrane-BoundImperfect modeling of structureStructures may change at bindingAnd even then…Molecule Binds Target But May:Molecule Binds Target But May:Bind too tightly or not tightly enough.Be toxic.Have other effects (side-effects) in the body.Break down as soon as it gets into the body, or may not leave the body soon enough.It may not get to where it should in the body (e.g., crossing blood-brain barrier).Not diffuse from gut to bloodstream.And Every Body is Different:And Every Body is Different:Even if a molecule works in the test tube and works in animal studies, it may not work in people (will fail in clinical trials).A molecule may work for some people but not others.A molecule may cause harmful side-effects in some people but not others.OutlineOutlineOverview of Drug DesignHow Machine Learning Fits Into the ProcessTarget Search: Single Nucleotide Polymorphisms (SNPs)Machine Learning from Feature VectorsDecision TreesSupport Vector MachinesVoting/EnsemblesPredicting Molecular Activity: Learning from StructurePlaces to use Machine LearningPlaces to use Machine LearningFinding target proteins.Inferring target site structure.Predicting who will respond positively/negatively.Places to use Machine LearningPlaces to use Machine LearningFinding target proteins.Inferring target site structure.Predicting who will respond positively/negatively.DiseasedHealthy Healthy vs. Disease Healthy vs. DiseaseIf We Could Sequence DNA Quickly and Cheaply, We Could:If We Could Sequence DNA Quickly and Cheaply, We Could:Sequence DNA of people taking a drug, and use ML to identify consistent differences between those who respond well and those who do not.Sequence DNA of cancer cells and healthy cells, and use ML to detect dangerous mutations… proteins these genes code for may be useful targets.Sequence DNA of people who get a disease and those who don’t, and use ML to determine genes related to succeptibility… proteins these genes code for may be useful targets.Problem: Can’t Sequence QuicklyProblem: Can’t Sequence QuicklyCan quickly test single positions where variation is common: Single Nucleotide Polymorphisms (SNPs).Can quickly test degree to which every gene is being transcribed: Gene Expression Microarrays (e.g., Affymetrix Gene Chips™).Can (moderately) quickly test which proteins are present in a sample (Proteomics).OutlineOutlineOverview of Drug DesignHow Machine Learning Fits Into the ProcessTarget Search: Single Nucleotide Polymorphisms (SNPs)Machine Learning from Feature VectorsDecision TreesSupport Vector MachinesVoting/EnsemblesPredicting Molecular Activity: Learning from StructureExample of SNP DataExample of SNP Data Person SNP 1 2 3 . . . CLASS Person 1 C T A G T T . . . old Person 2 C C A G C T . . . young Person 3 T T A A C C . . . old Person 4 C T G G T T . . . young . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Problem: SNPs are not GenesProblem: SNPs are not GenesIf we find a predictive SNP, it may not be part of a gene… we can only infer that the SNP is “near” a gene that may be involved in the disease.Even if the SNP is part of a gene, it may be another nearby gene that is the key gene.Problem: Even SNPs are CostlyProblem: Even SNPs are CostlyTypically cannot use all known SNPs.Can focus on a particular chromosome and area if knowledge permits that.Can use a scattering of SNPs, since SNPs that are very close together may be redundant… use one SNP per haplotype block, or region where recombination is