Review of class 4 what kinds of tests does 23andMe offer? significance and evaluation of each type risk – how do they know? where do the numbers come from? how believable are they? purpose of GWAS paper was to give you a feel for the answer to these kinds of questions
“Inverting” probabilities | given some condition Example: you observe freq of some genotype is higher in disease group vs controls, p(genotype|disease); but you want to know p(disease|genotype) Bayesian statistics tells you how to do the inversion Basic Idea: 2 ways to calculate p of disease and genotype AA p(D|AA)p(AA) = p(AA|D)p(D) -> p(D|AA) = p(AA|D) p(D) / p(AA) have to estimatep(D), p(AA) might estimate p(AA)@p(AA|control) if dis. rare if you just want relative risk, p(D|AA)/p(D) @ p(AA|D)/p(AA|control)
What did you learn from…. Math – correction to p value for large # of tests NEJM commentaries on GWAS Venter
NYT article on behavioral resp. to DTC genetic tests how similar were your responses? Should you have the right to keep the info conf.? from insurance co.? moral hazard? do you have the right? GINA ? pot. effect on classical insur. market – if risk prediction were better ? need for regulation, non-classical insur.
Evaluating value of pharmacogenetic tests Reduce hosp. – what are adv./disadv. of hist. controls? Cost-effectiveness study: gene test for warf. for 69 yo w/AF make tree of possible outcomes with prob. and costs til die disabl. p(1) embolism while waiting for stable dose no embolism p(2)bleed while establishing stable dose disabl.die no bleed w/gene test, assume p(1) lower by _ (dose estab. 2 d earl.) p(2) lower by 32% in gene var. (bec. less overdosing)
Total # life years saved x quality factors and costs /person Results – in group simulated to get gene test, # (qual adj.) life years saved/person = .0026 = 1d /person over expected life of 69yo gene test cost $367 cost per QALY = $367/.0026 = $144,000 some use $50,000/QALY as rough cut-off what gene test price would meet this cut-off? Note lots of assumptions, > 100 parameters in model Sensitivity analysis – how do results change if parameters are changed
Usual criticisms– So many assumptions! Results depend on assumptions What’s so special about $50K? What’s a good way (e.g. as an engineer!) to take uncertainties into account?
Next topic Impact of genome sequencing Goals – learn just enough “details” to be able to consider questions like: what is it good for? what kind of problems might it cause? how can we prepare for such problems? should aspects be regulated? genome “anatomy” consequences of diff. types of sequence variation how sequence is determined cost error rates frequency of diff. types of variants challenges in interpretation
Genome “anatomy” Size: ~3x109bp …..ACAGGTGCCCCATGA…… ……TGTCCACGGGGTACT….. Bases “pair” in dsDNA like zipper with 4 colors You can melt strands apart pieces with matching seq. find each other You can add “next” base at break to match templ.
Function of most of genome not understood Up to 50% “selfish” (? parasitic) DNA ~3% involved in regulating gene expression ~ 1.5% codes for protein (strings of amino acids) ~25,000 different proteins Basic process DNA->RNA (U replaces T) (some segments removed = “splicing”) RNA “read” into protein via complicated process that uses triplet code GAC AUG GGG CAG GAU GAT… UGA… mRNA met gly gln asp glu stop protein
Consequences of some mutations GAC AUG GGG CAG GAU GAT… UGA… wild type met gly gln asp glu stop GAC AUG GGG CAG UAU GAT… UGA SNP met gly gln tyr glu stop 1 aa subst. GAC AUG GGG C^GA GUA UGA T… UGA 1b insertion met gly arg valstopchanges all aa’s after GAC AUG GGG CCC CAG GAU GAT… UGA…3b inser. met gly pro gln asp glu stop1 aa inser. Frame-shifting “Indels” (non-mult of 3b) can have large effect
How can 23andMe determine ~106 SNPS from your spit? Array lots of ~3mm plastic beads into wells etched in fiberoptic bundle Each bead has a lots of copies of a single short DNA sequence; they know which bead is where; there are 106 different bead types! “Hybridize” your DNA to the array; add one base to the end of the bead DNA to copy your “template” – see if it is A, C, G, or T
How is sequence determined? several new methods “seq. by synthesis” – break genomic DNA into short pieces and array them on glass (may have to make many copies of each to increase signal); use enzyme that copies DNA 1 base at a time, using 4 dye-labeled nucleotides (e.g. A-red, C-green, .. etc) with blocking groups that prevent more than 1 base being added; color microscope image shows which base added to each DNA spot; then blocking group removed; next base added,…; can do ~108 different DNA templates on glass slide, getting ~30 bases/template; takes ~week to complete 1 genome
False color image of glass surface at one step in seq. by synthesis method; each spot = DNA template to which, e.g., A-red, C-green, G-blue or T-yellow has been added corresponding to complementary base in the template – Nature 456:53 (2008) Repeat ~ 30 times -> short bits of sequence to map to ref but at 108 different spots -> lots of sequence!
Costs Note how much and how quickly have they come down… Caveat: these are just reagent costs….
Error rates – how to measure them? Compare seq. to results of SNP chip hybridization ~99.9% accurate – but then how many errors/genome? Other kinds of errors - ~ 5% of seq. missed, too repetitive
What kind of variants are identified by sequencing? 3 individual’s genomes sequenced * They estimate ~10% of variants are false positives
Challenges to clinical interpretation Many variants are new, hence no clinical experience Even considering only those most likely to have biological effect (e.g. frameshifts in coding seq.), there will be hundreds of mutations per person Mutations may have no clinical effect because: if heterozygous, other allele provides enough protein; gene is non-essential; other proteins do same thing; gene function is only important in special circumstances Mutations with no effect in heterozygous parent could affect offspring that inherit 2 variant copies
Suppose one of the genomes sequenced was yours… What would you want to know? What would you be inclined to do?
If typical individual carries ~200 frame-shifting (FS) mut’ns, how likely is it that a couple will have at least one gene in which both carry FS mutations? p(FS mutation/gene) ~200/40,000 = 0.5% p(no shared gene with FS mutation) = .995200 = .37 p(>1 gene in which both are carriers) = 1-.37 = .63 What is probability that a fetus conceived by 2 people carrying FS mutations in the same gene will get both FS mutations? 1/4 What does this mean? Do ~16% of us (.63/4) have FS mutations in both copies of 1 or more genes?
If ~16% of us do have such double mutations, what would this say about those genes? If <<16% of us do, what might be the explanation? If you know you carry FS-mutations in 200 genes, would you want your potential mate to be tested to see if he/she carries mutations in the same genes? If you and your mate carried such mutations in the same genes, would you want prenatal testing?
What would you do if you found that a fetus you conceived carried 2 such mutations in the same gene? What would you want to know? How you think a physician/genetic counselor might respond? How would that make you feel?
Near-term uses of hi-throughput sequencing the genetically self-curious “next generation” GWAS prenatal diagnosis – small amount of fetal dna circulates in mother’s blood (along with mother’s DNA); is it easy to distinguish them?; how can hi thru-put sequencing be used to determine if fetus carries too many or too few copies of particular chromosomes (e.g. Down’s syn.)? cancer medicine – some mutations in cancer cells predict response or lack of resp. to various drugs
Clinical costs of whole genome sequencing - not just cost of initial sequencing confirmatory sequencing of potentially signif. variants ? testing of reproductive partners to check for overlapping mutations extra screening tests if results thought to predispose to screenable diseases Will sequencing save money or increase health-care costs? How can this be evaluated?
Should sequencing be regulated? If/when results are reasonably likely to have significant medical consequences? e.g. abortion, surgery, taking of medication, additional screening tests If you think seq. should be regulated in some circumstances, what should be goal? Certify accuracy? Disclosure of limitations in accuracy, interpretability? Require physician “prescription”? Have proven clinical utility?
Main points Cost decrease and technical feasibility suggest wide-spread genome sequencing is imminent Will identify large # of mutations per person Vast majority will be of uncertain significance Likely will advance understanding of disease mech. Raises host of social issues: regulatory, cost- effectiveness, data management (will genome seq. become part of medical record?); legal – who will have access to info?
Divide and conquer homework– pick 1 Greeley – get ready for fetal genome testing Chapter from F. Collins’ The Language of Life ch 3 Is it time to learn your own secrets? ch 4 Getting personal with the big C ch 5 What’s race got to do with it? ch 8 Genes and aging ch 9 The right drug at the right dose… Exercise on your response to your (hypothetical) genome sequence Another of the articles on Blackboard
Think about a student project – here are some ideas How does CMS decide what to cover? What laws prevent taking cost into consid.? Why are these laws there? What’s the debate? Your opinion? Ways forward – thinking outside of the box article by Pearson and Bach Limitations of cost-effectiveness (CE) studies ways to incorporate uncertainty (math) How much does US spend each year on low CE rx? article by Neumann et al
Potential incentives for reduced consumption of low CE rx Negative - e.g. higher co-pays Positive – e.g. payment to forego exp. Rx ethicalpros and cons what are potential consequences? Uncertainty in interpretation of sequence data Is this a problem for those tested? What are prospects for reduced uncertainty? How can consumers be adequately informed? What are likely consequences of the availability of this kind of information?