Scouting (autonomous exploration) </HEAD> <BODY> <CENTER> <H2>Autonomous Experimentation for Systems Biology</H2> </CENTER> <OL> <LI><A HREF="#intro">Towards Applied System Biology</A> <LI><A HREF="#scouting">Scouting Algorithm</A> <LI><A HREF="#application">Applications</A> <UL> <LI><A HREF="#app-vitro">In Vitro</A> <LI><A HREF="#app-silico">In Silico</A> <LI><A HREF="#app-ode">Deterministic Model</A> </UL> </OL> <A NAME="intro"><H3>Towards Applied Systems Biology</H3></A> Quantitative, predictive understanding of biology requires comprehensive information. High-throughput methods and laboratory automation technology have the potential to deliver the necessary data density. To harvest this potential, however, experimental design has to become dynamic and adaptive. Together with three application examples to demonstrate the versatility, we present a novel algorithm for <em>autonomously developing empirical models</em> by conducting experiments. One advantage of this technique (called scouting) is that no a priori knowledge about the systems is required. Perplexing complexity originating from intricate and manifold interplay of components is, therefore, preserved. Predicting systems behavior is a primary application of the empirically constructed <em>data-driven model</em>. Such a model replete with quantitative details can substitute preconceived mechanism; understanding follows application. <A NAME="scouting"><H3>Scouting Algorithm</H3></A> <UL> <LI><em>Evolutionary experimentation for exploration</em> <LI><em>Test outcomes immediately affect what test is performed next.</em> </UL> The scouting algorithm combines evolutionary computing with information theory to concentrate resources on areas of the experimentation space where prediction and observation deviate. An experiment is conducted according to experiment specification x generated by an evolution process and the observation r will be obtained. At the same time, an expectation r' based on the experience database is derived. The surprise value is the deviation between r and r', which is assigned to x as its fitness value. Observations are stored in the experience database so that in the end, it constitutes an empirical model. <IMG SRC=figs/scouting-fig-01.gif> <A NAME="application"><H3>Applications</H3></A> <UL> <LI><A NAME=app-vitro><H4>In Vitro: enzyme response surface</H4></A> The effect of divalent cations on the activity of malate dehydrogenase (MDH) is measured by monitoring NADH absorption over time. Enzyme activity analysis signals Products concentration L-Malate NADH Oxalacetate H+ NAD+ Mg Citrate 2+ MDH Concentration of Mg2+ and citrate is used as a chemical signal to be controlled by the scouting algorithm, and NADH absorption is observed as enzyme response to the signals. The experience database constitutes an empirical model of the enzyme s activity modulation. <BR> <IMG SRC=figs/overview-A-04.jpg>  <BR> <table border = "0"> <tr> <td><IMG height=300 SRC=figs/FluidicsFront02mod02.jpg></td> <td><IMG height=300 SRC=figs/FluidicsSide03mod02.jpg></td> </tr> </table> <IMG SRC=figs/MDH-response.gif> <LI><A NAME="app-silico"><H4>In Silico: chemical signature</H4></A> Coordinates Chemical composition An artificial planetary sphere has been implemented with a cellular automaton, simulating chemical dynamics. Biota introduce inhomogeneity into the system. When coordinates are specified, the chemical substance composition in this area is returned as the observation. The scouting algorithm has been successfully used to autonomously localize unusual chemical signatures. <table border = "0"> <tr> <td><IMG height=350 SRC=figs/sphere.gif></td> <td><IMG height=350 SRC=figs/sphere-cluster.gif></td> </tr> </table> <LI><A NAME="app-ode"><H4>Deterministic Model: HIV immunology</H4></A> A deterministic model of the relationship between HIV and the immune system was constructed by Wodarz and Nowak (1999). The scouting algorithm specifies initial states of the model, and the model returns the final states. 2000 observations in the experience database of scouting are analyzed. Surprise values are represented by colors. When the CTL precursors and effectors are chosen as the axis of the graph, a regular pattern was discovered, and this matches to the border of two behaviors of this model (i.e., virus is defeated and the immune system is destroy). <BR> <IMG height=250 SRC=figs/HIV-model.jpg> <BR> [Photo credit:© NIAID and AVERT] <BR> <BR><BR> <IMG HEIGHT=400 SRC=figs/xgobi-screen-02.gif><IMG HEIGHT=400 SRC=figs/xgobi-screen-color.gif> <BR> <BR><BR> <IMG HEIGHT=300 SRC=figs/HIV-analysis.gif> </UL> </BODY> </HTML>