APM 541: Stochastic Modelling in Biology Probability Notes Jay Taylor Fall 2013 JayTaylor (ASU) APM541 Fall2013 1/77 ; Outline Outline 1 Motivation 2 Probability and Uncertainty 3 Conditional Probability 4 Discrete Distributions 5 Continuous Distributions 6 Multivariate Distributions 7 Limit Theorems for Sums of Random Variables JayTaylor (ASU) APM541 Fall2013 2/77 Motivation Chance and Contingency Influence Biological Processes at all Scales quantum biology: stochastic tunneling in photosynthetic pathways stochastic molecular dynamics: molecular motors as Brownian ratchets noise in biochemical and gene regulatory networks developmental biology: stochastic X-inactivation leads to mosaic phenotypes in female mammals (tortoiseshell cats) behavioral ecology: optimal foraging in uncertain environments population ecology: extinction risk of small populations depends on demographic and environmental stochasticity population genetics: allele frequencies fluctuate because of random genetic drift and random mutation phylogenetics: random DNA substitutions are used for phylogenetic inference JayTaylor (ASU) APM541 Fall2013 3/77 Motivation Stochastic X-inactivation produces the Mass extinctions have had profound and mottled coat pattern on tortoiseshell often unpredictable consequences for life cats. on earth. JayTaylor (ASU) APM541 Fall2013 4/77 ProbabilityandUncertainty Multiple Interpretations of Probability Frequentist interpretation (J. Venn): The probability of an event is equal to its limiting frequency in an infinite series of independent, identical trials. Propensity interpretation (C. Pierce, K. Popper): The probability of an event is equal to the physical propensity of the event. (Confession: I have no idea what this means.) Logical interpretation (J. M. Keynes, R. Cox, E. Jaynes): Probabilities quantify the degree of support that some evidence E gives to a hypothesis H. Subjective interpretation (F. Ramsey, J. von Neumann, B. de Finetti): The probability of a proposition is equal to the degree of an individual’s belief that the proposition is true. JayTaylor (ASU) APM541 Fall2013 5/77 ProbabilityandUncertainty A Behavioristic Derivation of the Laws of Probability We consider a two-player game, with players P and P , that make wagers on an event 1 2 with a then-unknown outcome. Here we will assume that the event can result in one of n mutually-exclusive and exhaustive outcomes, e ,··· ,e . The game takes place in 1 n three steps. 1 First, player P1 assigns a unit price Pr(A) to each possible event A⊂S = {e ,··· ,e }. This is the amount that P is willing to pay for a $1 wager on the 1 n 1 event that A happens, as well as the amount that P is willing to accept from P 1 2 in exchange for such a wager. The player selling such a wager agrees to pay the purchaser $1 if A occurs and $0 if it does not occur. 2 PlayerP2 thendecidestheamountS(A)thattheywishtowageroneacheventA. If S(A) is positive, then P purchases a wager worth $S for S(A)·Pr(A) from P , 2 1 while if S(A) is negative, then P sells such a wager to P for the same amount. 2 1 3 The outcome of the event is then determined and the two players settle their wagers. JayTaylor (ASU) APM541 Fall2013 6/77 ProbabilityandUncertainty Although player P may assign any set of prices to the various possible wagers, many 1 assignments can be considered irrational insofar as they lead to certain loss for P . 1 Indeed, a set of prices {Pr(A):A⊂S} and a set of wagers {S(A):A⊂S} is said to be a Dutch book if these will guarantee that P will earn a profit from P no matter 2 1 which outcome occurs. The following are examples of Dutch books. Case 1: Suppose that Pr(A)<0 for some event A. In this case, P can agree to 2 purchase a $1 wager from P at a cost of Pr(A), i.e., P will pay P an amount |Pr(A)| 1 1 2 to take such a wager. Then P will have earned 1+|Pr(A)| if the event A occurs and 2 will have earned |Pr(A)| if the event A does not occur. Thus P makes a profit no 2 matter what occurs and thus this is a Dutch book. Case 2: SupposethatPr(S)<1. Inthiscase, P canearnaprofitof1−Pr(S)>0by 2 purchasing a $1 wager on S. Since one of the events e ,··· ,e is certain to happen, it 1 n follows that S will occur no matter which event occurs so that P is certain to make a 2 profit of 1−Pr(S)>0. Similarly, if Pr(S)>1, then P should sell a $1 stake in S to 2 P and this will guarantee that P will earn a profit of Pr(S)−1>0. 1 2 JayTaylor (ASU) APM541 Fall2013 7/77 ProbabilityandUncertainty Case 3: Now suppose that A and B are disjoint subsets of S, i.e., AB ≡A∩B =∅, and that Pr(A∪B)<Pr(A)+Pr(B). Assume that P purchases from P a $1 wager 2 1 on the event A∪B and sells to P both a $1 wager on A and a $1 wager on B. Then 1 the profit/losses incurred by P are summarized in the following table: 2 A B A∪B -1 -1 1 total outcome AB¯ Pr(A)−1 Pr(B) 1−Pr(A∪B) Pr(A)+Pr(B)−Pr(A∪B) A¯B Pr(A) Pr(B)−1 1−Pr(A∪B) Pr(A)+Pr(B)−Pr(A∪B) A¯B¯ Pr(A) Pr(B) −Pr(A∪B) Pr(A)+Pr(B)−Pr(A∪B) Notice that we do not consider the outcome AB since we are assuming that A and B represent mutually exclusive events. It follows that P earns a profit no matter which 2 outcome occurs and so P has made Dutch book against P . Similarly, if Pr(A∪B) 2 1 >Pr(A)+Pr(B), then P can make a Dutch book by selling a $1 wager on A∪B and 2 purchasing a $1 wager on each event A and B. JayTaylor (ASU) APM541 Fall2013 8/77 ProbabilityandUncertainty The above examples show that a Dutch book can be made against P whenever the 1 prices assigned by P to events in S satisfy certain inequalities. This motivates the 1 following definition. The set of prices (Pr(A):A⊂S) is said to be coherent if it satisfies the following conditions: 1 Pr(A)≥0 for each A⊂S; 2 Pr(S)=1; 3 Pr(A∪B)=Pr(A)+Pr(B) whenever A and B are disjoint subsets of S. Thusourpreviousexamplesshowthatanysetofpricesthatisnotcoherentisvulnerable to a Dutch book. In fact, the converse is also true: under the rules of the above game, P cannot make a Dutch book against P if the prices assigned by P are coherent. 2 1 1 JayTaylor (ASU) APM541 Fall2013 9/77 ProbabilityandUncertainty Condition (3) implies that a coherent price assignment is finitely-additive. If A ,··· ,A are disjoint subsets of S, then 1 m m ! m [ X Pr A = Pr(A). i i i=1 i=1 Indeed, this follows by recursion on m since we can use (3) to peel off sets A ,A ,··· one by one from the union inside the price function: m m−1 m ! m−1 ! m−1 ! [ [ [ Pr A = Pr A ∪A =Pr A +Pr(A ) i i m i m i=1 i=1 i=1 m−2 ! [ = Pr A +Pr(A )+Pr(A ) i m−1 m i=1 ··· m X = Pr(A). i i=1 JayTaylor (ASU) APM541 Fall2013 10/77
Description: