Models mutants

Contribution. In this paper, we propose a methodology for mutation-based debugging of real-time systems combining model-based debugging, classical mutation testing and model-based mutation testing. Given a faulty DUT and a testcase that fails when executed on it, we can determine a set of model mutants that reflects the implemented fault on model level. 

Within Section 2, we describe the basic concept of our novel model-based mutation debugging approach. Section 3 illustrates a model of a car alarm system, which is used as a running example during this paper. In Section 4 we define our deterministic Input/Output Timed Automata. In Section 5, we explain model mutation and a set of mutation operators for TA models and describe the linkage between model mutants and corresponding implementation faults. We explain our notion of the timed input/output conformance relation tioco and show the equivalence between language inclusion and tioco conformance in Section 6. 

Via model mutation we can create model mutants representing possible implementation faults. In different filtering steps we can select a small subset of those mutants showing the same faulty behavior as the DUT. These mutants are therefore likely to represent the implemented fault and can be seen as mutant diagnoses for the faulty implementation. Since correct timing behavior gains more and more importance in safety critical domains, we decided to use timed automata [5] to model the specification. 

Mutant Generation First, we create a set of all model mutants our framework supports (Mutantl, Mutant2, Mutants, Mutantd in Figure 1). Details on the different supported mutation operators can be found in Section 5. 

Mutant Generation we produce the whole set of model mutants. For the CAS the total number of mutants is 296. 

Conformance checks are needed at two different stages of om method The first conformance check is done in the Mutation Analysis step, between the abstract failing test case and the model mutants. This is possible, because our abstract test cases are timed automata traces in sequential form, that can be seen as partial models of the specification. Since the initial test case only covers a certain part of the specification, a lot of the mutations will be placed at parts that are not reached by the test case. Yet since the test case fails on the faulty implementation, we are only interested in parts of the model covered by the test case. Hence we can disregard each mutant that conforms to the test case. 

We applied our method to each of the faulty implementations in a separate experiment In each experiment, we used the specification model shown in Figure 2, one of the faulty implementations and a random test case of length 50, generated from the model by our tool MoMuT TA. If the random test case passed on the faulty implementation, new test cases were generated imtil one failed on it. All experiments used the same model mutants, which were produced from the specification model by our tool chain. The total number of timed automata model mutants for the CAS is 296. 

However, using this information, one can discard all mutants with mutations in the armedOn signal, further reducing the 17 mutants to 11. This is however the first step that requires manual input, while the execution so far can be done automatically. Table 1 presents the implementation faults represented by the remaining mutant diagnoses. The bold row shows the model mutant representing the actual implementation fault, which could easily be found with this information. 

Our initial test case was produced randomly with a length of 50 and is able to kill our faulty implementation. The tioco - conformance check between the model mutants and the test case reduced the total amount of mutants to 127, taking 742 seconds. Hence, 169 mutants were disregarded because the test case did not cover any unspecified output on them. 

IP Boissel, WR Lee, SR Presnell, EE Cohen, HP Bunn. Erythropoietin stiaicture-function relationships. Mutant proteins that test a model of tertiary stiaicture. I Biol Chem 268 15983-15993, 1993. 

C Lee. Testing homology modeling on mutant proteins Pi edictmg stiaictural and thermodynamic effects m the Ala98 Val mutants of T4 lysozyme. Folding Des 1 1-12, 1995. 

Model building also predicts that the Ala 216 mutant would displace a water molecule at the bottom of the specificity pocket that in the wild type enzyme binds to the NH3 group of the substrate Lys side chain (Figure 11.12). The extra CH3 group of this mutant is not expected to disturb the binding of the Arg side chain. One would therefore expect that the Km for Lys... 

Asp 189 at the bottom of the substrate specificity pocket interacts with Lys and Arg side chains of the substrate, and this is the basis for the preferred cleavage sites of trypsin (see Figures 11.11 and 11.12). It is almost trivial to infer, from these observations, that a replacement of Asp 189 with Lys would produce a mutant that would prefer to cleave substrates adjacent to negatively charged residues, especially Asp. On a computer display, similar Asp-Lys interactions between enzyme and substrate can be modeled within the substrate specificity pocket but reversed compared with the wild-type enzyme. 

As these experiments with engineered mutants of trypsin prove, we still have far too little knowledge of the functional effects of single point mutations to be able to make accurate and comprehensive predictions of the properties of a point-mutant enzyme, even in the case of such well-characterized enzymes as the serine proteinases. Predictions of the properties of mutations using computer modeling are not infallible. Once produced, the mutant enzymes often exhibit properties that are entirely surprising, but they may be correspondingly informative. 

Figure 11.16 Substrate-assisted catalysis. Schematic diagram from model building of a substrate, NHa-Phe-Ala-His-Tyr-Gly-COOH (red), bound to the subtilisin mutant His 64-Ala. The diagram illustrates that the His residue of the substrate can occupy roughly the same position in this mutant as His 64 in wild-type subtilisin (see Figure 11.14) and thereby partly restore the catalytic triad.

Model building shows that the OH group of Thr in the mutant is too far away to provide such a hydrogen bond. The loss of this feature of the stabilization of the transition state thus reduces the rate by more than a thousandfold. 

T4 lysozyme has two such cavities in the hydrophobic core of its a helical domain. From a careful analysis of the side chains that form the walls of the cavities and from building models of different possible mutations, it was found that the best mutations to make would be Leu 133-Phe for one cavity and Ala 129-Val for the other. These specific mutants were chosen because the new side chains were hydrophobic and large enough to fill the cavities without making too close contacts with surrounding atoms. 

Figure 17.14 Model of evolved mutant from cephalosphorinase shuffling. The sequence of the most active cephalosporinase mutant was modeled using the crystal structure of the class C cephalosporinase from Enterobacter cloacae. The mutant and wild-type proteins were 63% identical. This chimeric protein contained portions from three of the starting genes, including Enterobacter (blue), Klebsiella (yellow), and Citrobacter (green), as well as 33 point mutations (red). (Courtesy of A. Crameri.)...

These results indicate that is it possible to change the fold of a protein by changing a restricted set of residues. They also confirm the validity of the rules for stability of helical folds that have been obtained by analysis of experimentally determined protein structures. One obvious impliction of this work is that it might be possible, by just changing a few residues in Janus, to design a mutant that flip-flops between a helical and p sheet structures. Such a polypeptide would be a very interesting model system for prions and other amyloid proteins. 

Studies on the oxygen activation mechanisms by new heme enzymes using hemoprotein mutants and synthetic heme models 96YGK1046. 

Kojima, S., et al. (1998). Fluorescent properties of model chromophores of tyrosine-66 substitute mutants of Aequorea green fluorescent protein (GFP). Tetrahedron Lett. 39 5239-5242. 

The choice of the particular upward pathway in the kinetic resolution of rac-19, that is, the specific order of choosing the sites in ISM, appeared arbitrary. Indeed, the pathway B C D F E, without utilizing A, was the first one that was chosen, and it led to a spectacular increase in enantioselectivity (Figure 2.15). The final mutant, characterized by nine mutations, displays a selectivity factor of E=115 in the model reaction [23]. This result is all the more remarkable in that only 20000 clones were screened, which means that no attempt was made to fully cover the defined protein sequence space. Indeed, relatively small libraries were screened. The results indicate the efficiency of iterative CASTing and its superiority over other strategies such as repeating cycles of epPCR. 

---