|Rate-independent constructs for chemical computation.|
|Jump to Full Text|
|PMID: 21738654 Owner: NLM Status: In-Data-Review|
|THIS PAPER PRESENTS A COLLECTION OF COMPUTATIONAL MODULES IMPLEMENTED WITH CHEMICAL REACTIONS: an inverter, an incrementer, a decrementer, a copier, a comparator, a multiplier, an exponentiator, a raise-to-a-power operation, and a logarithm in base two. Unlike previous schemes for chemical computation, this method produces designs that are dependent only on coarse rate categories for the reactions ("fast" vs. "slow"). Given such categories, the computation is exact and independent of the specific reaction rates. The designs are validated through stochastic simulations of the chemical kinetics.|
|Phillip Senum; Marc Riedel|
Related Documents :
|9156174 - Metric analysis of threshold spectral sensitivity in the honeybee.
16812374 - Behavior regulation and learned performance: some misapprehensions and disagreements.
21711024 - Modeling the nonradiative-decay rate of electronically-excited thioflavin t.
12011404 - Computational organization science: a new frontier.
9156174 - Metric analysis of threshold spectral sensitivity in the honeybee.
11923994 - Analysing longitudinal continuous quality of life data with dropout.
|Type: Journal Article Date: 2011-06-30|
|Title: PloS one Volume: 6 ISSN: 1932-6203 ISO Abbreviation: PLoS ONE Publication Date: 2011|
|Created Date: 2011-07-08 Completed Date: - Revised Date: -|
Medline Journal Info:
|Nlm Unique ID: 101285081 Medline TA: PLoS One Country: United States|
|Languages: eng Pagination: e21414 Citation Subset: IM|
|Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, Minnesota, United States of America.|
|APA/MLA Format Download EndNote Download BibTex|
Journal ID (nlm-ta): PLoS One
Journal ID (publisher-id): plos
Journal ID (pmc): plosone
Publisher: Public Library of Science, San Francisco, USA
Senum, Riedel. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Received Day: 24 Month: 11 Year: 2010
Accepted Day: 1 Month: 6 Year: 2011
collection publication date: Year: 2011
Electronic publication date: Day: 30 Month: 6 Year: 2011
Volume: 6 Issue: 6
E-location ID: e21414
PubMed Id: 21738654
Publisher Id: PONE-D-10-05709
|Rate-Independent Constructs for Chemical Computation Alternate Title:Constructs for Chemical Computation|
|Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, Minnesota, United States of America
|Center for Genomic Regulation, Spain
|Correspondence: * E-mail: firstname.lastname@example.org
Contributed by footnote: Conceived and designed the experiments: PS. Performed the experiments: PS. Analyzed the data: PS. Contributed reagents/materials/analysis tools: PS MR. Wrote the paper: PS MR.
The theory of reaction kinetics underpins our understanding of biological and chemical systems . It is a simple and elegant formalism: chemical reactions define rules according to which reactants form products; each rule fires at a rate that is proportional to the quantities of the corresponding reactants that are present. On the computational front, there has been a wealth of research into efficient methods for simulating chemical reactions, ranging from ordinary differential equations (ODEs)  to stochastic simulation . On the mathematical front, entirely new branches of theory have been developed to characterize the dynamics of chemical reaction networks .
Most of this work is from the vantage point of analysis: a set of chemical reaction exists, designed by nature and perhaps modified by human engineers; the objective is to understand and characterize its behavior. Comparatively little work has been done at a conceptual level in tackling the inverse problem of synthesis: how can one design a set of chemical reactions that implement specific behavior?
Of course, chemical engineers, genetic engineers and other practitioners strive to create novel functionality all the time. Generally, they begin with existing processes and pathways and modify these experimentally to achieve the desired new functionality , . In a sense, much of the theoretical work on the dynamics of chemical reactions also addresses the synthesis problem by delineating the range of behaviors that are possible. For instance, theoretical work has shown that fascinating oscillatory and chaotic behaviors can occur in chemical reaction networks , .
Perhaps the most profound theoretical observation is that chemical reaction networks are, in fact, computational processes: regardless of the complexity of the dynamics or the subtlety of the timing, such networks transform input quantities of chemical species into output quantities through simple primitive operations. The question of the computational power of chemical reactions has been considered by several authors. Magnasco demonstrated that chemical reactions can compute anything that digital circuits can compute . Soloveichik et al. demonstrated that chemical reactions are Turing Universal, meaning that they can compute anything that a computer algorithm can compute .
Such prior work considered the computational power of chemical reactions from a deductive point of view. This paper tackles the problem from an inductive point of view. We present a constructive method for designing specific computational modules: an inverter, an incrementer, a decrementer, a copier, a comparator, a multiplier, an exponentiator, a raise-to-a-power operation, and a logarithm in base two. This work builds upon our prior work that described constructs such as “for” and “while” loops  and signal processing operations such as filtering .
In contrast to previous work, our method produces designs that are dependent only on coarse rate categories for the reactions (e.g., “fast” and “slow”). It does not matter how fast any “fast” reaction is relative to another, or how slow any “slow” reaction is relative to another – only that “fast” reactions are fast relative to “slow” reactions. Specifically, suppose that we design a module that requires slow reactions and fast reactions. Any choice of reactions with kinetic rate constants and reactions with kinetic rate constants , where , for all , for all , will work.
The result of the computation is rate-independent in the sense that the formula of what is computed, say a logarithm, does not include any of the kinetic rate constants. We do not mean to imply that the rates do not matter. If the separation between “slow” and “fast” is not sufficiently large, then errors will occur. However, for a sufficiently large separation, the errors are small.
Indeed, the error that occurs as a function of the separation between “fast” and “slow” is our main criterion of goodness for our design. As Tables 1, 2, 3, 4, 5, 6 illustrate, our constructs perform remarkably well, computing with small errors for rate separations of 100 or 1,000 and vanishingly small errors for rate separations of 10,000. We validate our all of our designs through stochastic simulations of the chemical kinetics  using an open-source tool called Cain . (Details about Cain can be found in the section “Analysis”.)
We adopt the model of discrete, stochastic chemical kinetics , . Molecular quantities are whole numbers (i.e., non-negative integers). Reactions fire and alter these quantities by integer amounts. The reaction rates are proportional to (1) the quantities of the reacting molecular types; and (2) kinetic rate constants. As discussed above, all of our designs are formulated in terms of two coarse kinetic rate constant categories (“fast” and “slow”).
Consider the reactionis consumed, one of is produced, and one of is produced. (Accordingly, is called a reactant and and the products.) Consider what this reaction accomplishes from a computational standpoint. Suppose that it fires until all molecules of have been consumed. This results in quantities of and equal to the original quantity of , and a new quantity of equal to zero:
Consider the reactionor all molecules of have been consumed. This results in a quantity of equal to the lesser of the two original quantities:
We will present constructs different arithmetical and logical operations in this vein. Each sets the final quantity of some molecular type as a function of the initial quantities of other types.
Most of our designs consist of either unimolecular or bimolecular reactions, i.e., reactions with one or two reactants, respectively. A small subset of the reactions are trimolecular. Mapping these to chemical substrates might not be feasible, since the kinetics of reactions with more than two reactants are complex and often physically unrealistic. For all trimolecular reactions, we suggest the follow generic scheme for converting them into bimolecular reactions. (This idea is found in  in the context of DNA strand displacement reactions.) We convert any trimolecular reactionis an new intermediary type. Note that Reaction 4 is a reversible reaction. We assume that this reaction is fast relative to all others. Accordingly, if there are non-zero quantities of and but zero of , the system will “back-off”, converting back into and . Other reactions in the system that use and can continue to fire.
Apart from reactions resulting from such trimolecular conversions, we do not use reversible reactions in any of our constructs. Of course, all chemical reactions are reversible. Implicitly, we assume that all reverse rates are much slower that the forward reactions (except for those corresponding to Reaction 4).
In this section, we present a collection of constituent constructs for rate-independent computation: an inverter, an incrementer/decrementer, a copier, and a comparator. In the next section, we use some of these constructs to implement a multiplier, a logarithm operation, an exponentiator, and a raise-to-the-power operation. A reference of all reactions needed for these constructs can be found in Supporting Information S1.
We implement an operation that is analogous to that performed by an inverter (i.e., a NOT gate) in a digital system: given a non-zero quantity (corresponding to logical “1”) we produce a zero quantity (corresponding to logical “0”). Conversely, given a zero quantity, we produce a non-zero quantity. We accomplish this with a pair of chemical types: the given type, for example, , and a corresponding “absence indicator” type, which will be referred to as . The reactions generating the absence indicator are shown in reactions 6–8., is used as a reactant, it indicates that the reactants are a large or replenishable unreactive source; when it is used as a product, it indicates that the products of the reaction are waste.
The first reaction continuously generates molecules of , so in the absence of molecules of we will have a non-zero quantity of in the system. If there are molecules of present, then second reaction quickly consumes any molecules of that are generated, so the quantity of will be close to zero. The third reaction ensures that the quantity remains small.
We use this simple construct in many of our computational modules , . It is also a fundamental part of all of the constructs introduced in this paper. In general, it can be used to synchronize steps. Suppose that we want to perform an operation similar to the one in reactions 9–10.. We do not want to start consuming molecules of until the full quantity of it is generated from . We can accomplish this with an absence indicator :
It is important to note that absence indicators generated by reactions 6–8 can only be used with “slow” reactions. If they were used by a “fast” reaction, it is possible that a false positive could be detected because a “fast” will compete with reaction 7. In situations where absence indicators need to be consumed by “fast” reactions, we can use an alternate two-step process to produce them.is produced from through a “slow” reaction. This allows “fast” reactions to use safely because it is buffered through reaction 16.
We describe constructs to implement incrementation and decrementation. These operations form the basis of more complex arithmetical operations, such as multiplication. The inputs consist of two molecular types: , the “start signal;” and , the quantity to be incremented or decremented. We assume that some external source injects molecules of . Any quantity can be injected; regardless, the quantity of is incremented or decremented by exactly one, consuming all the molecules in the process. The operations proceed as follows:
- The system waits for the start signal to be some non-zero quantity.
- It transfers the quantity of to a temporary type .
- It sets to zero.
- It transfers all but one molecule of back to .
- For a decrement, it removes the last molecule .
- For an increment, it removes the last molecule of and adds to two molecules to .
The following reactions implement this scheme. Given molecules of , a reaction transfers molecules of to molecules of :to zero. Using the absence indicator mechanism described in the preceding section, it does so only once all molecules of have been transfered to : ; we omit them here. The following reaction transfers all but one molecule of back to . It does so by repeatedly selecting pairs of and turning one molecule of into . In essence, this is a repeated integer division by two. Again, using the absence indicator mechanism, it proceeds only once all molecules of have been removed: , but instead, we use a secondary absence indicator , generated in the method outlined in reactions 13–18.
Reaction 21 also produces molecules of a supplementary type . Note that this reaction is in the “fast” category. The new type is consumed by the reaction:because we cannot directly use an absence indicator for to detect when reaction 21 has completed because is never completely consumed. Instead, we use to indicate that we are currently transferring molecules of back to ; it is consumed when the step completes. Again, reactions of the same form as 6–8 are needed to generate ; we omit them here.
Finally, we include the following reaction to perform a decrement:
In digital computation, one of the most basic operations is copying a quantity from one register into another. The programming construct is “set the value of to be the value of ”:to : , setting its quantity to zero. We would like a chemical construct that copies the quantity without altering it. The following reaction does not work either: in the presence of . A more sophisticated construct is needed.
In our construct, we have a “start signal” type . When an external source injects molecules of , the copy operation proceeds. (In the same way as our increment and decrement operations, the quantity of that is injected is irrelevant.) It produces an output quantity of equal to the input quantity of ; it leaves the quantity of unchanged.
The reactions for the copier construct are as follows. Firstly, in the presence of , a reaction transfers the quantity of to :have been transferred to , the system removes all the molecules of : ensures that is copied exactly once.
After has been removed, a reaction transfers the quantity of back to and also creates this same quantity of ::
We also generate absence indicators and by the same method as reactions 6–8.
We note that, while this construct leaves the quantity of unchanged after it has finished executing, it temporarily consumes molecules , transferring the quantity of these to before transferring it back. Accordingly, no other constructs should use in the interim.
Using our copier construct, we can create a construct that compares the quantities of two input types and produces an output type if one is greater than the other. For example, let us assume that we want to compare the quantities of two types and :
If the quantity of is greater than the quantity of , the system should produce molecules of an output type ; otherwise, it should not produce any molecules of .
First, we create temporary copies, and , of the types that we wish to compare, and , using the copier construct described in the previous section. (We omit these reactions; they are two verbatim copies of the copier construct, one with as an input and as an output, the other with as an input and as an output.) We split the start signal so that the two copiers are not competing for it:
Now we compare and via their respective copies and . To start, we first consume pairs of and :left, or only molecules of left, or no molecules of nor left. Molecules of the type that originally had a larger quantity have persisted. If the quantities were equal, then both types were annihilated. We use absence indicators and to determine which type was annihilated, produced by the method shown in reactions 6–8. If was originally greater than , there will now be a presence of and an absence of . We produce molecules of type if this condition is met. We preserve the quantities of and . We can also limit the quantity of produced by introducing a fuel type: in the case that the asserted condition is not true:Table 1 lists these operations and their corresponding reactions.
Based upon the modules described in the previous section, we provide examples of how to implement more complex arithmetic: multiplication, logarithms in base two, exponentiation, and raising to a power. In order to elucidate the designs, we specify the sequence of operations for each of these module in pseudo-code. The pseudo-code operations consist of:
- Assignment, addition, and subtraction operations. The operands may be constants or variables.
- Decision-making constructs: while and if statements. The logical test for each of these constructs can be any one of the six conditions listed in Table 1. In some cases, the if and while statements will be nested.
Building upon the constructs in the last section, we show a construct that multiplies the quantities of two input types. Multiplication can be implemented via iterative addition. Consider the following lines of pseudo-code:is equal to times . We implement multiplication chemically using the constructs described in the previous sections: the line z = z+y is implemented with a copy operation; the line x = x−1 is implemented using a decrement operation. A third set of reactions handle the looping behavior of the while statement. The reactions presented here are also listed in Supporting Information S2.
Firstly, we have reactions that copy the quantity of to . We use “start signal” types and to synchronize iterations; it is supplied from the controlling reaction 48 below.. We use as the signal to begin the decrement. and to begin the next iteration, preserving the quantity of : in the system. The types and are present when we are decrementing or copying , respectively; thus, they can be used to decide whether we are currently inside the loop or not. Finally, we generate the four absence indicators according to the template in reactions 6–8.
As a second complex example, we show how to implement the operation . This can be done using iterative multiplication; as we demonstrated in the previous section, multiplication can be implemented via iterative addition. The pseudo-code for the raising-to-a-power operation is shown in Figure 1. It consists of assignment, addition, subtraction, and iterative constructs. Note that the assignment operations can be performed with our “copier” module; the addition and subtraction operations can be performed with “increment” and “decrement” modules. A pair of nested while constructs, similar to that used for multiplication, perform the requisite iterative computation. The complete set of reactions to implement this operation is given in Supporting Information S3.
To implement the operation , we can use a sequence of operations similar to those that we used for multiplication. The pseudo-code is shown in Figure 2. The reactions that implement this pseudo-code are given in Supporting Information S4.
We demonstrate the computation of a base-2 logarithm. The pseudo-code is shown in Figure 3. A logarithm is the inverse operation of exponentiation; it makes sense, therefore, that the pseudo-code for is more or less the reverse of that for exponentiation. The reactions that implement this operation are given in Supporting Information S5.
We validated our constructs using stochastic simulation. Specifically, we performed a time homogeneous simulation using Gillespie's “Direct Method”  with the software package “Cain” from Caltech . (Details about Cain can bew found in the section “Analysis”.) In each case, the simulation was run until the quantities of all types except the absence indicators converged to a steady state. We used a rate constant of 1 for the “slow” reactions. We tried rate constants between two to four orders of magnitude higher for the “fast” reactions. (We refer to the ratio of “fast” to “slow” as the rate separation.) For each of the graphs below, the initial quantity of each type is zero, with the exception of the types specified.
Figure 4 shows the output of a single simulated trajectory for our multiplier. We observe exactly the behavior that we are looking for: the quantity of cycles exactly 10 times as it exchanges with and is copied to ; the quantity of grows steadily up to 100; the quantity of decreases once each cycle down to 0. Table 1 presents detailed simulation results, this time tested for accuracy. Errors generally occur if the system executes too many or too few iterations. As can be seen, the larger the quantity of , the more accurate the result, in relative terms. As expected, the larger the rate separation, the fewer errors we get.
Figure 5 shows an average simulated trajectory for our copier. Again, we observe exactly the behavior we expect: the quantity of drops to 0 almost immediately as it turns into ; this is followed by the removal of from the system. When the quantity of drops to nearly zero, both and rise steadily back to the original quantity of . Table 2 shows additional simulation results from our copier, this time tested for accuracy. The copier construct appears to be quite robust to errors; however, large rate separations do not help as much as they do for the multiplier. The system seems to prefer a larger injection quantity of , but whether it is larger or smaller than the initial quantity of is irrelevant.
Figure 6 shows the output of a single simulated trajectory of our decrementer. An automatic restart mechanism, similar to reactions 45–48, was used to produce a continuous series of decrements. Exactly twenty peaks can be seen in the graph, including the initial peak on the far-left margin of the graph. This is exactly the behavior we are looking for – a decrement by exactly one each cycle.
Figure 9 shows a simulated trajectory of our raise-to-a-power construct. As can be observed, after is loaded with the initial quantity of , it is multiplied by twice. Each time its value is stored in the temporary type before being transferred back. Table 4 shows simulation results for our raise-to-a-power construct for various values of and . In each case, the initial quantity of was set to 10, simulating an injection of that type.
Figure 10 shows a simulated trajectory of our exponentiation construct. We can observe that for every decrementation of , doubles in value, which is the behavior that we are aiming for. Table 5 shows more simulation results. The error for this construct is small but appears to grow as grows. This is not surprising, given we are performing exponentiation: small errors will be compounded.
This paper presented a collection of specific computational constructs. More complex operations – multiplication, exponentiation, raising to a power, and logarithms – were built a collection of robust, primitive operations – absence indicators, incrementing and decrementing, copying, and comparing. The process by which we assembled these primitive operations could be readily generalized. Indeed, we are developing a chemical compiler that will translate any sequence of operations specified by pseudo-code into chemical reactions. The compiler will accept general pseudo-code written in the vein of that shown in Figures 1, 2, 3. It will allow for assignments, arithmetic operations, “if” statements, and arbitrarily nested “while” loops.
The novelty and value of the constructs that we have demonstrated is that they are all rate independent. Here “rate independent” refers to the fact that, within a broad range of values for the kinetic constants, the computation does not depend on the specific values of the constants. Of course, outside of this range, the accuracy of the computation degrades. For rates within the target range, our results are remarkably accurate: in nearly all cases the errors are less than 1%. In many cases, the errors are much less than 1%. The actual value of the target range will depend on the chemical substrate used; in simulation, it was found that a ratio of 10,000∶1 of “fast” vs. “slow” produced nearly perfect results.
Our contribution is to tackle the problem of synthesizing computation at a conceptual level, working not with actual molecular types but rather with abstract types. One might question whether actual chemical reactions matching our templates can be found. Certainly, engineering complex new reaction mechanisms in any experimental domain is formidable task; for in vivo systems, there are likely to be many experimental constraints on the choice of reactions . However, we point to recent work on in vitro computation as a potential application domain for our ideas.
Through a mechanism called DNA strand-displacement, a group at Caltech has shown that DNA reactions can emulate the chemical kinetics of nearly any chemical reaction network. They also provide a compiler that translates abstract chemical reactions of the sort that we design into specific DNA reactions . Recent work has demonstrated both the scale of computation that is possible with DNA-based computing , as well as exciting applications . While conceptual, our work suggest a de novo approach to the design of biological functions. Potentially this approach is more general in its applicability than methods based on appropriating and reusing existing biological modules.
All the models described in this paper are contained in an XML file. This file is available at:
The file is designed for use with Cain, a biochemical simulator from Caltech . It contains initial quantities for all types. All non-zero quantities can be modified as the user desires to simulate different input values. Within Cain, we suggest Gillespie's Direct Method for all simulations.
Appendix: Module Reference.
Click here for additional data file (pone.0021414.s001.pdf)
Supporting Information S2
Appendix: Multiplication Reactions.
Click here for additional data file (pone.0021414.s002.pdf)
Supporting Information S3
Appendix: Raise-to-a-Power Reactions.
Click here for additional data file (pone.0021414.s003.pdf)
Supporting Information S4
Appendix: Exponentiation Reactions.
Click here for additional data file (pone.0021414.s004.pdf)
Supporting Information S5
Appendix: Logarithm Reactions.
Click here for additional data file (pone.0021414.s005.pdf)
Competing Interests: The authors have declared that no competing interests exist.
Funding: This research is supported by an NSF CAREER Award, #0845650 (http://www.nsf.gov/awardsearch/showAward.do?AwardNumber=0845650&WT.z_pims_id), and by a UROP Award from the University of Minnesota (http://www.research.umn.edu/undergraduate/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
|1.||Horn F,Jackson R. Year: 1972General mass action kinetics.Archive for Rational Mechanics and Analysis4781116|
|2.||Érdi P,Tóth J. Year: 1989Mathematical Models of Chemical Reactions: Theory and Applications of Deterministic and Stochastic ModelsManchester University Press|
|3.||Gillespie D. Year: 1977Exact stochastic simulation of coupled chemical reactions.Journal of Physical Chemistry8123402361|
|4.||Strogatz S. Year: 1994Nonlinear Dynamics and Chaos with Applications to Physics, Biology, Chemistry, and Engineering.Perseus Books|
|5.||Win MN,Liang J,Smolke CD. Year: 2009Frameworks for programming biological function through RNA parts and devices.Chemistry & Biology1629831019318211|
|6.||Keasling J. Year: 2008Synthetic biology for synthetic chemistry.ACS Chemical Biology3647618205292|
|7.||Epstein IR,Pojman JA. Year: 1998An Introduction to Nonlinear Chemical Dynamics: Oscillations, Waves, Patterns, and ChaosOxford Univ Press|
|8.||Willamowski KD,Rössler OE. Year: 1980Irregular oscillations in a realistic abstract quadratic mass action system.Zeitschrift fur Naturforschung Section A – A Journal of Physical Sciences35317318|
|9.||Magnasco MO. Year: 1997Chemical kinetics is turing universal.Phys Rev Lett7811901193|
|10.||Soloveichik D,Cook M,Winfree E,Bruck J. Year: 2008Computation with finite stochastic chemical reaction networks.Natural Computing7|
|11.||Shea A,Fett B,Riedel MD,Parhi K. Year: 2010Writing and compiling code into biochemistry.Proceedings of the Pacific Symposium on Biocomputing456464|
|12.||Jiang H,Kharam AP,Riedel MD,Parhi KK. Year: 2010A synthesis ow for digital signal processing with biomolecular reactions.IEEE International Conference on Computer-Aided Design417424|
|13.||Gillespie DT. Year: 1976A general method for numerically simulating the stochastic time evolution of coupled chemical reactions.Journal of Computational Physics22403434|
|14.||Mauch S,Stalzer M. Year: 2009Efficient formulations for exact stochastic simulation of chemical systems. IEEE/ACM Transactions on Computational Biology and Bioinformatics 99.|
|15.||Gillespie DT. Year: 2006Stochastic simulation of chemical kinetics.Annual Review of Physical Chemistry583555|
|16.||Soloveichik D,Seelig G,Winfree E. Year: 2010DNA as a universal substrate for chemical kinetics.Proceedings of the National Academy of Sciences10753935398|
|17.||Kharam A,Jiang H,Riedel MD,Parhi K. Year: 2011Binary counting with chemical reactions.Pacific Symposium on Biocomputing|
|18.||Weiss R. Year: 2003Cellular Computation and Communications using Engineering Genetic Regulatory Networks. Ph.D. thesis, MIT.|
|19.||Qian L,Winfree E. Year: 2009A simple DNA gate motif for synthesizing large-scale circuits.DNA Computing7089|
|20.||Venkataramana S,Dirks RM,Ueda CT,Pierce NA. Year: 2010 (in press)Selective cell death mediated by small conditional RNAs. Proceedings of the National Academy of Sciences.|
Table 1 Logical operations via chemical reactions.
|a = = b||a> = b|
|a>b||a< = b|
|a<b||a! = b|
Table 2 Statistical simulation results for “Multiplier” construct.
Table 3 Statistical simulation results for “Copier” construct.
Table 4 Statistical Simulation Results from “Raise to a Power” Construct.
Table 5 Statistical Simulation Results from “Exponentiation” Construct.
Table 6 Statistical Simulation Results from “Logarithm” Construct.
Previous Document: Effect of commercially available egg cures on the survival of juvenile salmonids.
Next Document: Berberine Inhibits Human Hepatoma Cell Invasion without Cytotoxicity in Healthy Hepatocytes.