PDF《upon Mutation》

0 then it costs more in energy to have the wild type structure than the mutant one thus the mutation is more favorable to the structure stability. Conversely, if ΔΔG >

0, the mutant structure ΔG is higher than the wild type one thus the mutation is less favorable to the structure stability. Evaluation of the Stability of Folding Nucleus upon Mutation

57 2.4 Stability Analysis Many methods have been implemented to predict stability changes induced by point mutations. MUpro [23] and I-Mutant (sequence version) [24] both predict stability changes on a protein sequence whereas DFIRE [25], I-Mutant (sequence + structure version) [24] and PoPMuSiC [26] use protein sequence and structure to predict these changes. Other methods exist but have been rejected due to some restrictions: CUPSAT1 [27] was not available in a standalone version and the cur- rent version of FoldX2 [28] only computes mutations to Alanine. To avoid biases from one or the other method, we present a comprehensive analysis with ?ve ex- isting tools (the two versions of I-Mutant are considered as two di?erent tools). We use the Protherm database [29] that collects thermodynamic data pub- lished in the scienti?c literature and thus includes measured values of ΔΔG to compare our prediction to experimental data. It is available at the following URL: http://gibk26.bse.kyutech.ac.jp/jouhou/Protherm/protherm.html. 2.5 Data Set Our analysis was conducted on a dataset published by the Protein Folding Frag- ments European consortium that can be found at the URL: http://bioserv.rpbs. univ-paris-diderot.fr/PFF/. MIR predictions and TEF calculations were already performed on the selected

116 protein sequences. The experimental dataset consisted of

116 protein sequences for a total of 15,183 amino acids. Each sequence was processed with each of the ?ve sta- bility prediction tools, for each amino acid, and for each of the

19 possible mutations. We computed 1,442,385 di?erent ΔΔG values. In order to man- age and publish our produced data and results in a more e?cient way than output ?at ?les, a database was created and is available to the community at http://bioinformatics.eas.asu.edu/Stability/ where more information about the data can be found.

3 Results 3.1 MIR and TEF The MIR concept aims at characterizing the main amino acids involved in the early steps of the protein folding process. The TEF method splits a structure into fragments with spatially close ends that interact with each other. A previous study [10] has demonstrated that TEF ends (within a range of ±5 positions) correspond to MIR in

57 % of the cases. As we are looking at coherent methods to determine the folding nucleus, we observe that MIR over predict the TEF ends, therefore we hypothesize that restricting the MIR to the ones in agreement with TEF ends would capture the expected amino acids responsible for protein folding.

1 CUPSAT is available at http://cupsat.tu-bs.de/.

2 FoldX is available at http://foldx.crg.es//.

58 M. Lonquety, Z. Lacroix, and J. Chomilier Then, separation between good MIR and bad MIR emerges, and we de- ?ne them as TEF related and TEF independent MIR. The TEF independent MIR are expected to be the noise in the MIR prediction algorithm. The TEF related ones ar........