Lecture outline. Optimal Experimental Design: Where to find basic information. Theory of D-optimal design

Similar documents
Numerical Enzymology

Chapter 11. Supplemental Text Material. The method of steepest ascent can be derived as follows. Suppose that we have fit a firstorder

Team. Outline. Statistics and Art: Sampling, Response Error, Mixed Models, Missing Data, and Inference

Additional File 1 - Detailed explanation of the expression level CPD

Specification -- Assumptions of the Simple Classical Linear Regression Model (CLRM) 1. Introduction

Improvements on Waring s Problem

2.3 Least-Square regressions

Scattering of two identical particles in the center-of. of-mass frame. (b)

No! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Survey Results. Class 15. Is the following possible?

Not at Steady State! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Yes! Class 15. Is the following possible?

MULTIPLE REGRESSION ANALYSIS For the Case of Two Regressors

Introduction. Modeling Data. Approach. Quality of Fit. Likelihood. Probabilistic Approach

Small signal analysis

Improvements on Waring s Problem

Statistical Properties of the OLS Coefficient Estimators. 1. Introduction

NAME (pinyin/italian)... MATRICULATION NUMBER... SIGNATURE

AP Statistics Ch 3 Examining Relationships

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281

Start Point and Trajectory Analysis for the Minimal Time System Design Algorithm

Introduction to Interfacial Segregation. Xiaozhe Zhang 10/02/2015

Harmonic oscillator approximation

Supporting Information. Hydroxyl Radical Production by H 2 O 2 -Mediated. Conditions

The Electric Potential Energy

HOMEWORK ASSIGNMENT #2

Department of Mechanical Engineering Massachusetts Institute of Technology Modeling, Dynamics and Control III Spring 2002

is the calculated value of the dependent variable at point i. The best parameters have values that minimize the squares of the errors

Design By Emulation (Indirect Method)

Variable Structure Control ~ Basics

Estimation of Finite Population Total under PPS Sampling in Presence of Extra Auxiliary Information

Massachusetts Institute of Technology Dynamics and Control II

A Simplified Methodology for the Synthesis of Adaptive Flight Control Systems

Chapter 6 The Effect of the GPS Systematic Errors on Deformation Parameters

S_LOOP: SINGLE-LOOP FEEDBACK CONTROL SYSTEM ANALYSIS

Gain and Phase Margins Based Delay Dependent Stability Analysis of Two- Area LFC System with Communication Delays

The multivariate Gaussian probability density function for random vector X (X 1,,X ) T. diagonal term of, denoted

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS

Bogoliubov Transformation in Classical Mechanics

CHAPTER 4 DESIGN OF STATE FEEDBACK CONTROLLERS AND STATE OBSERVERS USING REDUCED ORDER MODEL

This appendix presents the derivations and proofs omitted from the main text.

Given the following circuit with unknown initial capacitor voltage v(0): X(s) Immediately, we know that the transfer function H(s) is

POWER SYSTEM SMALL SIGNAL STABILITY ANALYSIS BASED ON TEST SIGNAL

GNSS Solutions: What is the carrier phase measurement? How is it generated in GNSS receivers? Simply put, the carrier phase

Alpha Risk of Taguchi Method with L 18 Array for NTB Type QCH by Simulation

Chapter 11: Simple Linear Regression and Correlation

Root Locus Techniques

Modeling of Wave Behavior of Substrate Noise Coupling for Mixed-Signal IC Design

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

Lecture 17: Analytic Functions and Integrals (See Chapter 14 in Boas)

Midterm Review - Part 1

On the SO 2 Problem in Thermal Power Plants. 2.Two-steps chemical absorption modeling

CHAPTER 9 LINEAR MOMENTUM, IMPULSE AND COLLISIONS

3 Implementation and validation of analysis methods

Correction for Simple System Example and Notes on Laplace Transforms / Deviation Variables ECHE 550 Fall 2002

Two Approaches to Proving. Goldbach s Conjecture

Electrical Circuits II (ECE233b)

Singular perturbation theory

Solution Methods for Time-indexed MIP Models for Chemical Production Scheduling

BOUNDARY ELEMENT METHODS FOR VIBRATION PROBLEMS. Ashok D. Belegundu Professor of Mechanical Engineering Penn State University

A Hybrid Evolution Algorithm with Application Based on Chaos Genetic Algorithm and Particle Swarm Optimization

Batch RL Via Least Squares Policy Iteration

Linear Momentum. calculate the momentum of an object solve problems involving the conservation of momentum. Labs, Activities & Demonstrations:

8 Waves in Uniform Magnetized Media

Alternate Dispersion Measures in Replicated Factorial Experiments

Social Studies 201 Notes for November 14, 2003

PHYS 450 Spring semester Lecture 02: Dealing with Experimental Uncertainties. Ron Reifenberger Birck Nanotechnology Center Purdue University

Social Studies 201 Notes for March 18, 2005

Lecture 8: Period Finding: Simon s Problem over Z N

Information Acquisition in Global Games of Regime Change (Online Appendix)

Assignment for Mathematics for Economists Fall 2016

Introduction to Laplace Transform Techniques in Circuit Analysis

Hybrid Projective Dislocated Synchronization of Liu Chaotic System Based on Parameters Identification

Originated from experimental optimization where measurements are very noisy Approximation can be actually more accurate than

Computer Control Systems

Image Registration for a Series of Chest Radiograph Images

General Topology of a single stage microwave amplifier

DIFFERENTIAL EQUATIONS

This is a repository copy of An iterative orthogonal forward regression algorithm.

Chapter 4. The Laplace Transform Method

Math 273 Solutions to Review Problems for Exam 1

online learning Unit Workbook 4 RLC Transients

5.5 Application of Frequency Response: Signal Filters

Wind - Induced Vibration Control of Long - Span Bridges by Multiple Tuned Mass Dampers

Simple Observer Based Synchronization of Lorenz System with Parametric Uncertainty

Lecture 16 Statistical Analysis in Biomaterials Research (Part II)

Pythagorean triples. Leen Noordzij.

ME 375 FINAL EXAM Wednesday, May 6, 2009

1 Routh Array: 15 points

Linear Motion, Speed & Velocity

into a discrete time function. Recall that the table of Laplace/z-transforms is constructed by (i) selecting to get

arxiv: v1 [cs.gt] 15 Jan 2019

Estimating floor acceleration in nonlinear multi-story moment-resisting frames

Lecture 10 Filtering: Applied Concepts

Electromagnetic scattering. Graduate Course Electrical Engineering (Communications) 1 st Semester, Sharif University of Technology

Control Systems Analysis and Design by the Root-Locus Method

A Preliminary Study on Material Utilization of Stiffened Cylindrical Shells

Physics 741 Graduate Quantum Mechanics 1 Solutions to Final Exam, Fall 2014

The Influence of the Load Condition upon the Radial Distribution of Electromagnetic Vibration and Noise in a Three-Phase Squirrel-Cage Induction Motor

Lectures on Multivariable Feedback Control

Confidence intervals for the difference and the ratio of Lognormal means with bounded parameters

Optimal Coordination of Samples in Business Surveys

Transcription:

v I N N O V A T I O N L E C T U R E (I N N O l E C) Lecture outlne Bndng and Knetc for Expermental Bologt Lecture 8 Optmal degn of experment The problem: How hould we plan an experment uch we learn the mot from t? Petr Kuzmč, Ph.D. BoKn, Ltd. WATERTOWN, MAACHUETT, U..A. The oluton: Ue the Optmal Degn Theory of tattc An mplementaton: oftware Dynat An example: Knetc of clathrn cage daembly BKEB Lec 8: Optmal degn Optmal Expermental Degn: Where to fnd bac nformaton DOZEN O BOOK Theory of D-optmal degn MAXIMIZE THE DETERMINANT ( D ) O IHER INORMATION MATRIX edorov, V.V. (97) Theory of Optmal Experment edorov, V.V. & Hacl, P. (997) Model-Orented Degn of Experment Atnon, A.C & Donev, A.N. (99) Optmum Expermental Degn Endreny, L., Ed. (98) Degn and Analy of Enzyme and Pharmaconetc Experment y f (, p) f algebrac fttng functon x x ndependent varable, th data pont (,,, N) y dependent varable, th data pont p vector of M model parameter f ( x, p) p, entvty of f wth repect to th parameter, th data pont, N,,,,, (,)th element of the, her nformaton matrx M M, D-Optmal Degn:, L, M, L, M O M M, L M, M M max x, x,, xn det Chooe the ndependent varable x,, x N (e.g., total or ntal concentraton of reagent) uch that the determnant of maxmzed. BKEB Lec 8: Optmal degn 3 BKEB Lec 8: Optmal degn 4 D-Optmal degn example: Mchael-Menten equaton RONALD DUGGLEBY UNIVERITY O QUEENLAND, AUTRALIA (979) J. Theor. Bol. 8, 67-684 (979) v V + K v, V V + K v ntal rate of enzyme reacton, th data pont ubtrate concentraton, th data pont (,,, N) V, K vector of model parameter K Mchael contant, V maxmum rate v V K ( + ), K K entvty functon Realtc degn for the Mchael-Menten equaton ININITE UBTRATE CONCENTRATION (TO GET V max) I IMPOIBLE TO ACHIEVE Box-Luca two-pont degn wth one pont ( max) already gven: aume K 0.5, max.0 max det max, V, V max, K, K.0 Box-Luca two-pont degn: max det,, V, V K, K, K.0 K 0.5 V.0 0. 0.0 0 3 BKEB Lec 8: Optmal degn 5 max max K max + K RECIPE In the determnaton of K M, alway nclude a ubtrate concentraton that correpond to a reacton rate approxmately one half of maxmum achevable rate. v 0. 0.0 0 3 0.3333 BKEB Lec 8: Optmal degn 6

Theory of D-optmal degn: The chcen-and-egg problem PROBLEM: TO DEIGN AN EXPERIMENT WE MUT IRT GUE THE INAL ANWER! EQUENTIAL OPTIMAL DEIGN perform a mallet poble experment refne parameter etmate repeat pecal cae: Tme-coure experment wth fxed tme-pont TART HERE gue model parameter (lterature, hunche, ) optmal degn for very few data pont optmal degn theory ue here BKEB Lec 8: Optmal degn 7 BKEB Lec 8: Optmal degn 8 A typcal netc experment: xed meh of tme-pont PROBLEM: WE DO NOT ALWAY HAVE A CHOICE O INDEPENDENT VARIABLE VALUE Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35 topped flow expermentaton: xed tep ze AT BET, WE CAN CHANGE THE (IXED) TEP IZE OR A GIVEN INTERVAL Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35, g. 4. lght catterng..0 0.9 0.7 [Hc70] µm lght catterng.5.0.05.00 0.95 0.90 5 0 0.00 0.0 0.04 0.06 0.08 0.0 tme, ec t 0 to 5 ec: tep ze 0 mec 0.5 0 4 6 8 0 tme, ec lght catterng.5.0.05.00 0.95 0.90 5 0 5.00 5.0 5.04 5.06 5.08 5.0 tme, ec t 5 to 0 ec: tep ze 00 mec BKEB Lec 8: Optmal degn 9 BKEB Lec 8: Optmal degn 0 Varaton of the degn problem: Optmze ntal condton I WE CAN T CHOOE OBERVATION TIME, AT LEAT WE CAN CHOOE INITIAL CONCENTRATION Theory of D-optmal degn: Intal condton n ODE ytem MAXIMIZE THE DETERMINANT ( D ) O IHER INORMATION MATRIX BAIC PRINCIPLE: In netc tude -thendependent varable tme; - ntal concentraton are condered parameter of the model. D-Optmal degn theory concerned wth optmal choce of ndependent varable - n th cae the obervaton tme. Unfortunately, n the real-world we cannot chooe partcular obervaton tme: - uual ntrument are offerng u only a fxed meh of output pont. But we can turn thng around and - treat ntal concentraton a ndependent varable. - Then we can optmze the choce of ntal concentraton, ung the uual formalm of the D-Optmal Degn theory. dc / dt f ( c,) t 0 : c c 0 ntal value problem (frt-order ordnary dfferental equaton) c vector of concentraton vector of rate contant c 0 concentraton at tme zero y g( c t ),r) y expermental gnal at th data pont (tme t ), N ( g( c (, t ), r), p,, c r, concentraton at tme t vector of molar repone and/or offet on gnal ax entvty of f wth repect to th parameter, th data pont p model parameter: vector and r combned D-Optmal Degn: max det c 0 BKEB Lec 8: Optmal degn BKEB Lec 8: Optmal degn

Optmze ntal condton: Dynat notaton THE OTWARE TAKE CARE O ALL THE MATH Optmze ntal condton: Algorthm and Dynat ettng THE DIERENTIAL EVOLUTION ALGORITHM REQUIRE PECIAL ETTING ta degn [mechanm] yntax otherwe ued for confdence nterval [data] et concentraton X?? (0.0.. 00) th value gnored (preent for yntactcal reaon only) lower and upper bound mut be gven ta degn copy thee ettng from one of the [mechanm] dtrbuted example problem [ettng] {DfferentalEvoluton} Populatonzexed 300 populaton not too large MaxmumEvoluton MnmumEvoluton perform the optmzaton only once TetParameterRange TetParameterRangeAll TetParameterRangeull topparameterrange 0. TetCotunctonRange y relatvely wea convergence crtera topcotunctonrange 0.0 TetCotunctonChange y topcotunctonchange 0.0000 TetCotunctonChangeCount 5 BKEB Lec 8: Optmal degn 3 BKEB Lec 8: Optmal degn 4 Clathrn tructure: trelon and cage CLATHRIN CAGE ARE LARGE ENOUGH TO VIIBLE IN MICROCOPY AND LIGHT CATTERING Cae tudy: Knetc of clathrn cage daembly clathrn trelon clathrn cage BKEB Lec 8: Optmal degn 5 BKEB Lec 8: Optmal degn 6 Clathrn bology: Role n endocyto CLATHRIN I INVOLVED IN INTRACELLULAR TRAICKING Eenberg et al. (007) Traffc 8, 640-646 In vtro netc of clathrn daembly: Expermental data WATCHING CLATHRIN CAGE TO ALL APART BY PERPENDICULAR LIGHT CATTERING Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35, g. 4 µm Hc70 ATP-dependent uncoatng clathrn-coated vecle Clathrn cage (0.09 μm trela) premxed wth 0. μm auxln were mxed wth Hc70 (concentraton n μm hown on graph) and 500 μm ATP, and perpendcular lght catterng wa meaured ung topped-flow BKEB Lec 8: Optmal degn 7 BKEB Lec 8: Optmal degn 8 3

In vtro netc of clathrn daembly: Theoretcal model MODEL ELECTION UING THE AKAIKE INORMATION CRITERION (DYNAIT) Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35 In vtro netc of clathrn daembly: Dynat notaton THE MOT PLAUIBLE MODEL: THREE TEP EQUENTIAL Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35 THREE-TEP EQUENTIAL: DYNAIT INPUT: AUTOMATICALLY GENERATED MATH MODEL: TWO-TEP EQUENTIAL: THREE-TEP CONCERTED: ta ft model AHAHAH? [mechanm] CA + T > CAT CAT > CAD + P : r CAD + T > CADT CADT > CADD + P : r CADD + T > CADDT CADDT > CADDD + P : r CADDD > Prod : d ta ft model AHAH? Etc. In total fve dfferent model were evaluated. BKEB Lec 8: Optmal degn 9 BKEB Lec 8: Optmal degn 0 In vtro netc of clathrn daembly: Preferred mechanm CONCLUION: THREE ATP MOLECULE MUT BE HYDROLYZED BEORE THE CAGE ALL APART Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35 In vtro netc of clathrn daembly: Raw data THI WA A VERY EXPENIVE EXPERIMENT TO PERORM Rothne, Kuzmc, et al. (0) Proc. Natl. Acad. c. UA 08, 697-35 ACTUAL EXPERIMENTAL DATA: 0.5 0 [Hc70] x aay / experment 90 nm clathrn 00 nm auxln 4 0.5 up to 4 µm Hc70 a lot of materal expenve and/or tme conumng to obtan BKEB Lec 8: Optmal degn BKEB Lec 8: Optmal degn How many aay are actually needed? D-OPTIMAL DEIGN IN DYNAIT ta degn [mechanm] CA + T > CAT CAT > CAD + P CAD + T > CADT CADT > CADD + P CADD + T > CADDT CADDT > CADDD + P CADDD > Prod : r : r : r : d [contant] a 9? r 6.5? d 0.38? Chooe eght ntal concentraton of T uch that the rate contant a, r, d are determned mot precely. [data] fle run0 concentraton CA 0., T?? (0.00.. 00) fle run0 concentraton CA 0., T?? (0.00.. 00) fle run03 concentraton CA 0., T?? (0.00.. 00) fle run04 concentraton CA 0., T?? (0.00.. 00) fle run05 concentraton CA 0., T?? (0.00.. 00) fle run06 concentraton CA 0., T?? (0.00.. 00) fle run07 concentraton CA 0., T?? (0.00.. 00) fle run08 concentraton CA 0., T?? (0.00.. 00) Optmal Expermental Degn: Dynat reult URPRIE: WE DID TOO MUCH WORK OR THE INORMATION GAINED IMULATED DATA OPTIMAL EXPERIMENT: D-Optmal ntal concentraton: [T] 0.70 µm, 0.73 µm [T].4 µm,.5 µm,.5 µm [T] 76 µm, 8 µm, 90 µm maxmum feable concentraton upwng phae no longer een Jut three experment would be uffcent for follow-up! One half of the materal compared to the orgnal experment. BKEB Lec 8: Optmal degn 3 BKEB Lec 8: Optmal degn 4 4

Optmal Expermental Degn: Dynat reult - dcuon EACH O THE THREE UNIQUE AAY TELL A DIERENT TORY Optmal Expermental Degn n Dynat: ummary NOT A ILVER BULLET! motly ATP aocaton ( a) [mechanm] CA + T > CAT CAT > CAD + P : r CAD + T > CADT CADT > CADD + P : r CADD + T > CADDT CADDT > CADDD + P : r CADDD > Prod : d Ueful for follow-up (verfcaton) experment only - Mechantc model mut be nown already - Parameter etmate mut alo be nown Tae a very long tme to compute - Contraned global optmzaton: Dfferental Evoluton algorthm - Clathrn degn too 30-90 mnute - Many degn problem tae multple hour of computaton Crtcally depend on aumpton about varance motly daembly ( d) aocaton ( upwng ) no longer vble BKEB Lec 8: Optmal degn 5 - Uually we aume contant varance ( noe ) of the gnal - Mut verfy th by plottng redual agant gnal (not the uual way) BKEB Lec 8: Optmal degn 6 5