# Novel Descriptors and Digital Signal Processing- Based Method for Protein Sequence Activity Relationship Study

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Results

#### 2.1. Concatenation to Obtain Extended Numerical Sequences

_{j1}, FFT_Seq

_{j2}… and noFFT_Seq

_{j1}, noFFT_Seq

_{j2}…, respectively.

_{j1}--noFFT_Seq

_{j2}

_{j1}--noFFT_Seq

_{j2}

_{j1}--FFT_Seq

_{j2}

_{j1}--FFT_Seq

_{j2}

_{j2}--FFT_Seq

_{j1}.

_{Nb_Index}which is strictly greater than two.

#### 2.2. Combining One or Multiple Indices

#### 2.2.1. Concatenation of the Best Single Indices with or/and without FFT

^{2}) drops from 0.85 with FFT to 0.64 without FFT and the root mean squared error in cross validation (cvRMSE) rises from 0.32 to 0.48, using only the best single index: Thus, applying FFT significantly improves the quality of the model [22].

^{2}and cvRMSE are respectively 0.83 and 1.91. The results obtained using this best index when an extended sequence noFFT

_{i1}-FFT

_{i1}generated are cvR

^{2}= 0.83 and cvRMSE = 1.89 (Figure 2b). The p-value associated with Student’s t-test used to assess if the difference for the quadratic errors between the two models indicates that this slight decrease in cvRMSE is not significant (p-value = 0.7781).

_{i1}-FFT_Seq

_{i2}(Figure 2c) for cytochrome P450. Here, in parallel, the best index was selected without FFT. Figure 2c shows the results obtained for the extended sequence noFFT

_{i2}-FFTi

_{1:}cvR

^{2}and cvRMSE to be respectively 0.84 and 1.85. p-value decreases but remains quite high in this case (p-value = 0.4123).

#### 2.2.2. Combinatorial of a Protein Sequence Encoded by Multiple Indices with FFT

#### Combinatorial Sequences Approach by Using a Selection of Best Single Indices

_{j1}--FFT_Seq

_{ji2}

_{j2}--FFT_Seq

_{j3}

_{j1}--FFT_Seq

_{j2}

^{_}FFT_Seq

_{j4}

_{j1}-- FFT_Seq

_{j2}--FFT_Seq

_{j3}--FFT_Seq

_{j4}--“FFT_Seq

_{j5}--FFT_Seq

_{j6}-- FFT_Seq

_{j7}--FFT_Seq

_{j8}--FFT_Seq

_{j9}-- FFT_Seq

_{j10}

**GLP-2**. We applied this approach to the GLP-2 dataset and from a previous study [22] the index 449 was shown as the best after a ranking of indices and encoding with FFT.

^{2}and cvRMSE are respectively 0.42 and 2.11 for the index 449.

_{1}-FFT_Seqj

_{2}is equivalent to FFT_Seqi

_{2}-FFT_Seq_i

_{1}, 175 combined extended sequences are obtained.

^{2}and cvRMSE with three indices are respectively 0.47 and 1.99, with p-value = 0.53. Thus, in this case the modeling performance appears better but the improvement is not significant according to the p-value (p-value = 0.531). Nevertheless, interesting findings were obtained. Indeed, we tested the model using the 10 indices to form the Ext_SEQ FFT

_{i1}-FFT

_{i2}….FFT

_{i10.}The cvRMSE of this model jumps to 2.48 and the cvR

^{2}decreases to 0.11. So, it should be noted that the right number of indices has to be found: i.e., a combination of m index is not always better than a combination of n index (with m > n). In other words, large Ext_SEQ is not equal to better modeling performances. The addition of indices for the encoding step is not related to the improvement of the modeling. Moreover, we notice that index 449, the best index when only one index is selected, could not be the best to use for a combination of three indices from the top ten indices as exemplified in Table 1. Indeed, 449 alone appears in position nine in the ranking.

**Epoxide Hydrolase**

^{2}, respectively. The best performances, seen in Table 2, are 0.105 and 0.969, respectively for cvRMSE and cvR

^{2}, with the p-value = 0.43. Thus, the combinatorial of multiple indices appears to slightly improve the modeling performances but the improvement is not significant.

#### Successive Concatenation of a Protein Sequence Encoded by Multiple Indices

_{1}, using a sequence representation noFFT_Seq or FFT_Seq. In the second iteration, the process identified another index, j2, to use for the construction of Ext_SEQ of two Ele_SEQ, starting from the sequence encoded by j1 as a base block of Ext_SEQ. The index j2 is identified by a second ranking with all the indices except the one used in the base block of Ext_SEQ, j1, for the second iteration, i.e., the ranking on 566—one indice. For each iteration the same operation is repeated to find the best index for modeling and increasing the size of the Ext_SEQ.

_{j1}-FFT_Seq

_{j2}-..-FFT_Seq

_{jn}is obtained. This could be extended to any number of parts in the “Ext_SEQ”. Furthermore, a mix of noFFT and FFT could be used.

**GLP-2 Dataset**

^{2}and cvRMSE, respectively (cf. Figure 3). Figure 6 shows the results obtained using the three indices, 449, 341, and 193, gathered in the Ext_SEQ “FFT_SEQ

_{j1}--FFT_Seq

_{j2}--FFT_Seq

_{j3}”. cvR

^{2}and cvRMSE are 0.55 and 1.75, respectively. Thus, using the three indices significantly improves the quality of the prediction. This is confirmed by the p-value equal to 0.008 in Student’s test for the significance of the improvement.

**Epoxide Hydrolase Dataset**

^{2}and the cvRMSE. Figure 7 resulted from the application of the successive concatenation method to the epoxide hydrolase dataset. The combination with the indices 14 and 234 gives better performances since 0.97 and 0.09 are respectively obtained for the cvR

^{2}and the cvRMSE but the improvement in comparison to one index is not significant, with a p-value of 0.343. Nevertheless, we note that here the p-value is lower than the value (0.43) shown in Table 2 above.

**TNF Alpha Dataset**

^{2}and cvRMSE, for the TNF dataset. The combination with the indices 504 and 486 (Figure 8b) allows increasing cvR

^{2}to 0.88 and decreasing cvRMSE to 0.28 (p-value = 0.175).

**Cytochrome P450 Dataset**

^{2}and 1.91 as cvRMSE. The combination with the indices 39 and 226 allows significant improvement of these performances, up to 0.88 and 1.63, respectively for the cvR

^{2}and the cvRMSE. This improvement is confirmed by the p-value (0.002).

#### 2.3. Selection of Indices from Different Families for Concatenation or Combination

- Alpha and turn propensities,
- beta propensity,
- composition,
- hydrophobicity,
- physicochemical properties,
- other properties.

#### 2.4. Selection of Variables Inside the Ext_SEQ

^{2}, and the set of frequencies is then given a percentage of frequencies for which R

^{2}is the highest. A given set of frequencies was selected and used for modeling.

_{20%}.

_{20%}_Seq with index number 300.

_{j1}-- FFT

_{20%}_Seq

_{j2}with j1 equal to index number 343 and j2 equal to index number 300.

_{20%}: CvR

^{2}and cvRMSE are 0.66 and 2.68, respectively. With the same encoding index 300, without FFT and with FFT

_{20%}, Figure 10b shows better results obtained with the prediction method: CvR

^{2}and cvRMSE are 0.74 and 2.38, respectively. With the two best encoding indices (index numbers 300 and 343) and FFT

_{20%}for index number 300, Figure 10c shows better results obtained with the prediction method: cvR

^{2}and cvRMSE are 0.74 and 2.39, respectively.

^{2}and cvRMSE are 0.70 and 2.52, respectively. In each case, the improvement is significant compared to the reference (see Table 4).

^{2}and cvRMSE values for Figure 10a–d.

- (i)
- (ii)

## 3. Discussion

#### 3.1. Cumulating Indices Could Provide Better Prediction Performances

^{2}. Table 4 sums up the performance metrics in LOOCV.

^{2}and cvRMSE from one model to another is significant or not, i.e., to know if the accumulation of indices, with or without FFT, produces an improvement. Different cases are observed with respect to the p-value associated with the difference for the quadratic errors between two models; they are summarized in Table 4:

^{2}= 0.83 and cvRMSE = 1.91 to 0.88 and 1.63, respectively for Ext_SEQ with three indices. Student’s t-test shows a significant improvement of performances with a p-value = 1.95 × 10

^{−3}.

#### 3.2. PLSR Modeling Using FFT and Extended Sequence Leads to Better Results

#### 3.3. Considerations on the Nature of the Descriptors and Versatility of the Approach

^{2}varying from 0.55 to 0.97 using the descriptor Ext_SEQ (with FFT). cAMP activation refers to the potency and to the measure of drug activity expressed in terms of the amount required to produce an effect of given intensity. Binding affinity refers to the strength of interactions between proteins or proteins and ligands (peptide or small chemical molecule). The thermostability is usually expressed in °C and usually refers to the measured activity T

_{50}defined as the temperature at which 50% of the protein is irreversibly denatured after an incubation time of 10 min. Finally, enantioselectivity refers to the selectivity of a reaction towards one of a pair of enantiomers.

#### 3.4. Calculation Time

## 4. Materials and Methods

#### 4.1. Datasets

#### 4.1.1. Epoxide Hydrolase

^{‡}(kcal/mol) by the relation $\mathsf{\Delta}\mathsf{\Delta}{G}^{\u2021}=-RTln\left(E\right)$. The modeling is based on the ΔΔG

^{‡}values.

#### 4.1.2. Cytochrome P450

_{50}. T

_{50}is the temperature at which 50% of the protein irreversibly denatured after incubation for 10 min. The values of T

_{50}range from 39.2 to 64.4 °C.

#### 4.1.3. GLP-2

#### 4.1.4. TNF Alpha

_{d}) of TNF to its two receptors, TNFR1 and TNFR2, is computed as a single ratio of log

_{10}(R1/R2) which ranges from 0 to 2.87, where R1 and R2 are affinities of TNF to TNFR1 and TNFR2, respectively as measured by IC

_{50}assays in ng/mL.

#### 4.2. Modeling Approach Based on Single or Multiple Encoding with or without Fast Fourier Transform (FFT)

_{j}|; the numerical sequence includes N value(s) denoted x

_{k}, with 0 ≤ k ≤ N − 1 and N ≥ 1; and i defines the imaginary number such that i² = −1.

^{2}) are the performance parameters to assess a regression model in the validation step. RMSE values vary between 0 and +∞. The R

^{2}value varies between 0 and 1. An accurate regression model has an RMSE close to 0 and a R

^{2}close to 1.

#### 4.3. Evaluation of Modeling Performances

^{2}during the cross-validation stage.

^{2}values. While R

^{2}is a measure of the extent of agreement between measured and predicted fitness, cvRMSE represents the extent to which the predictions vary when different training sets are used.

_{i}is the measured activity of the i

^{th}sequence, ŷ

_{i}is the predicted activity of the i

^{th}sequence, $\overline{y}$ is the average of measured activities, $\widehat{\overline{y}}$ is the average of the predicted activities and n the number of sequences.

#### 4.4. Measure of the Significance of the Improvement between Two Models

## 5. Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Damborsky, J.; Brezovsky, J. Computational tools for designing and engineering enzymes. Curr. Opin. Chem. Biol.
**2014**, 19, 8–16. [Google Scholar] [CrossRef] [PubMed] - Sumbalova, L.; Stourac, J.; Martinek, T.; Bednar, D.; Damborsky, J. HotSpot Wizard 3.0: Web server for automated design of mutations and smart libraries based on sequence input information. Nucleic Acids Res.
**2018**, 46, W356–W362. [Google Scholar] [CrossRef] [PubMed] - Romero-Rivera, A.; Garcia-Borràs, M.; Osuna, S. Computational tools for the evaluation of laboratory-engineered biocatalysts. Chem. Commun.
**2017**, 53, 284–297. [Google Scholar] [CrossRef] [PubMed] - Yang, K.K.; Wu, Z.; Bedbrook, C.N.; Arnold, F.H. Learned protein embeddings for machine learning. Bioinformatics
**2018**, 34, 2642–2648. [Google Scholar] [CrossRef] [PubMed] - Wu, Z.; Kan, S.B.J.; Lewis, R.D.; Wittmann, B.J.; Arnold, F.H. Machine learning-assisted directed protein evolution with combinatorial libraries. Proc. Natl. Acad. Sci. USA
**2019**, 116, 8852–8858. [Google Scholar] [CrossRef] [PubMed] - Li, Y.; Drummond, D.A.; Sawayama, A.M.; Snow, C.D.; Bloom, J.D.; Arnold, F.H. A diverse family of thermostable cytochrome P450s created by recombination of stabilizing fragments. Nat. Biotechnol.
**2007**, 25, 1051–1056. [Google Scholar] [CrossRef] [PubMed] - Hellberg, S.; Sjöström, M.; Wold, S. The Prediction of Bradykinin Potentiating Potency of Pentapeptides. An Example of a Peptide Quantitative Structure-activity Relationship. Acta Chem. Scand.
**1986**, 40, 135–140. [Google Scholar] [CrossRef] [PubMed] - Norinder, U.; Rivera, C.; Undén, A. A quantitative structure-activity relationship study of some substance P-related peptides a multivariate approach using PLS and variable selection. J. Pept. Res.
**2009**, 49, 155–162. [Google Scholar] [CrossRef] [PubMed] - Wold, S.; Trygg, J.; Berglund, A.; Antti, H. Some recent developments in PLS modeling. Chemom. Intell. Lab. Syst.
**2001**, 58, 131–150. [Google Scholar] [CrossRef] - Lapinsh, M.; Prusis, P.; Gutcaits, A.; Lundstedt, T.; Wikberg, J.E.S. Development of proteo-chemometrics: A novel technology for the analysis of drug-receptor interactions. Biochim. Acta BBA Gen. Subj.
**2001**, 1525, 180–190. [Google Scholar] [CrossRef] - Fox, R. Directed molecular evolution by machine learning and the influence of nonlinear interactions. J. Theor. Biol.
**2005**, 234, 187–199. [Google Scholar] [CrossRef] [PubMed] - Li, G.; Dong, Y.; Reetz, M.T. Can Machine Learning Revolutionize Directed Evolution of Selective Enzymes? Adv. Synth. Catal.
**2019**. [Google Scholar] [CrossRef] - Qu, G.; Li, A.; Sun, Z.; Acevedo-Rocha, C.G.; Reetz, M.T. The Crucial Role of Methodology Development in Directed Evolution of Selective Enzymes. Angew. Chem. Int. Ed.
**2019**. [Google Scholar] [CrossRef] [PubMed] - Berland, M.; Offmann, B.; Andre, I.; Remaud-Simeon, M.; Charton, P. A web-based tool for rational screening of mutants libraries using ProSAR. Protein Eng. Des. Sel.
**2014**, 27, 375–381. [Google Scholar] [CrossRef] [PubMed] - Smith, S.W. The Scientist and Engineer’s Guide to Digital Signal Processing; California Technical Publishing: San Diego, CA, USA, 1997. [Google Scholar]
- Cadet, X.F.; Dehak, R.; Chin, S.P.; Bessafi, M. Non-Linear Dynamics Analysis of Protein Sequences. Application to CYP450. Entropy
**2019**, 21, 852. [Google Scholar] [CrossRef] - Cosic, I. Macromolecular bioactivity: Is it resonant interaction between macromolecules?-theory and applications. IEEE Trans. Biomed. Eng.
**1994**, 41, 1101–1114. [Google Scholar] [CrossRef] [PubMed] - Walsh, I.; Sirocco, F.G.; Minervini, G.; Di Domenico, T.; Ferrari, C.; Tosatto, S.C.E. RAPHAEL: Recognition, periodicity and insertion assignment of solenoid protein structures. Bioinformatics
**2012**, 28, 3257–3264. [Google Scholar] [CrossRef] [PubMed] - Hrabe, T.; Godzik, A. ConSole: Using modularity of Contact maps to locate Solenoid domains in protein structures. BMC Bioinform.
**2014**, 15, 119. [Google Scholar] [CrossRef] [PubMed] - Nwankwo, N. Digital Signal Processing Techniques:Calculating Biological Functionalities. J. Proteom. Bioinform.
**2012**, 4. [Google Scholar] [CrossRef] - Jia, J.; Liu, Z.; Xiao, X.; Liu, B.; Chou, K.-C. iPPI-Esml: An ensemble classifier for identifying the interactions of proteins by incorporating their physicochemical properties and wavelet transforms into PseAAC. J. Biol.
**2015**, 377, 47–56. [Google Scholar] [CrossRef] [PubMed] - Cadet, F.; Fontaine, N.; Vetrivel, I.; Chong, M.N.F.; Savriama, O.; Cadet, X.; Charton, P. Application of fourier transform and proteochemometrics principles to protein engineering. BMC Bioinform.
**2018**, 19, 382. [Google Scholar] [CrossRef] [PubMed] - Cadet, F.; Fontaine, N.; Li, G.; Sanchis, J.; Chong, M.N.F.; Pandjaitan, R.; Vetrivel, I.; Offmann, B.; Reetz, M.T. A machine learning approach for reliable prediction of amino acid interactions and its application in the directed evolution of enantioselective enzymes. Sci. Rep.
**2018**, 8, 16757. [Google Scholar] [CrossRef] [PubMed] - Ostafe, R.; Fontaine, N.; Frank, D.; Ng Fuk Chong, M.; Prodanovic, R.; Pandjaitan, R.; Offmann, B.; Cadet, F.; Fischer, R. One-shot optimization of multiple enzyme parameters: Tailoring glucose oxidase for pH and electron mediators. Biotechnol. Bioeng.
**2019**. [Google Scholar] [CrossRef] [PubMed] - Prusis, P.; Lundstedt, T.; Wikberg, J.E.S. Proteo-chemometrics analysis of MSH peptide binding to melanocortin receptors. Protein Eng. Des. Sel.
**2002**, 15, 305–311. [Google Scholar] [CrossRef] [PubMed] - Barley, M.H.; Turner, N.J.; Goodacre, R. Improved Descriptors for the Quantitative Structure–Activity Relationship Modeling of Peptides and Proteins. J. Chem. Inf. Model.
**2018**, 58, 234–243. [Google Scholar] [CrossRef] [PubMed] - Kawashima, S.; Pokarowski, P.; Pokarowska, M.; Kolinski, A.; Katayama, T.; Kanehisa, M. AAindex: Amino acid index database, progress report 2008. Nucleic Acids Res.
**2007**, 36, D202–D205. [Google Scholar] [CrossRef] [PubMed] - Tomii, K.; Kanehisa, M. Analysis of amino acid indices and mutation matrices for sequence comparison and structure prediction of proteins. Protein Eng.
**1996**, 9, 27–36. [Google Scholar] [CrossRef] [PubMed] - Sneath, P.H.A. Relations between chemical structure and biological activity in peptides. J. Theor. Biol.
**1966**, 12, 157–195. [Google Scholar] [CrossRef] - Chou, P.Y.; Fasman, G.D. Prediction of the Secondary Structure of Proteins from Their Amino Acid Sequence. In Advances in Enzymology - and Related Areas of Molecular Biology; Meister, A., Ed.; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2006; pp. 45–148. [Google Scholar] [CrossRef]
- Palau, J.; Argos, P.; Puigdomenech, P. Protein Secondary Structure. Studies on the Limits of Prediction Accuracy. Int. J. Pept. Protein Res.
**1982**, 19, 394–401. [Google Scholar] [CrossRef] [PubMed] - Rackovsky, S.; Scheraga, H.A. Differential Geometry and Polymer Conformation. 4. Conformational and Nucleation Properties of Individual Amino Acids. Macromolecules
**1982**, 15, 1340–1346. [Google Scholar] [CrossRef] - Robson, B.; Suzuki, E. Conformational Properties of Amino Acid Residues in Globular Proteins. J. Mol. Biol.
**1976**, 107, 327–356. [Google Scholar] [CrossRef] - Naderi-Manesh, H.; Sadeghi, M.; Arab, S.; Moosavi Movahedi, A.A. Prediction of Protein Surface Accessibility with Information Theory. Proteins
**2001**, 42, 452–459. [Google Scholar] [CrossRef] - Bull, H.B.; Breese, K. Surface Tension of Amino Acid Solutions: A Hydrophobicity Scale of the Amino Acid Residues. Arch. Biochem. Biophys.
**1974**, 161, 665–670. [Google Scholar] [CrossRef] - Levitt, M. Conformational Preferences of Amino Acids in Globular Proteins. Biochemistry
**1978**, 17, 4277–4285. [Google Scholar] [CrossRef] [PubMed] - Meek, J.L. Prediction of Peptide Retention Times in High-Pressure Liquid Chromatography on the Basis of Amino Acid Composition. Proc. Natl. Acad. Sci. USA
**1980**, 77, 1632–1636. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Prabhakaran, M. The Distribution of Physical, Chemical and Conformational Properties in Signal and Nascent Peptides. Biochem. J.
**1990**, 269, 691–696. [Google Scholar] [CrossRef] [PubMed] [Green Version] - George, R.A.; Heringa, J. An Analysis of Protein Domain Linkers: Their Classification and Role in Protein Folding. Protein Eng.
**2002**, 15, 871–879. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Di Giulio, M. A Comparison of Proteins from Pyrococcus Furiosus and Pyrococcus Abyssi: Barophily in the Physicochemical Properties of Amino Acids and in the Genetic Code. Gene
**2005**, 346, 1–6. [Google Scholar] [CrossRef] [PubMed] - Nakashima, H.; Nishikawa, K.; Ooi, T. Distinct Character in Hydrophobicity of Amino Acid Compositions of Mitochondrial Proteins. Proteins
**1990**, 8, 173–178. [Google Scholar] [CrossRef] [PubMed] - Kumar, S.; Tsai, C.J.; Nussinov, R. Factors Enhancing Protein Thermostability. Protein Eng.
**2000**, 13, 179–191. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Nakashima, H.; Nishikawa, K. The Amino Acid Composition Is Different between the Cytoplasmic and Extracellular Sides in Membrane Proteins. FEBS Lett.
**1992**, 303, 141–146. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Rackovsky, S.; Scheraga, H.A. Hydrophobicity, Hydrophilicity, and the Radial and Orientational Distributions of Residues in Native Proteins. Proc. Natl. Acad. Sci. USA
**1977**, 74, 5248–5251. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Reetz, M.T.; Sanchis, J. Constructing and Analyzing the Fitness Landscape of an Experimental Evolutionary Process. ChemBioChem
**2008**, 9, 2260–2267. [Google Scholar] [CrossRef] [PubMed] - Iakovou, K.; Kazanis, M.; Vavayannis, A.; Bruni, G.; Romeo, M.R.; Massarelli, P.; Teramoto, S.; Fujiki, H.; Mori, T. Synthesis of oxypropanolamine derivatives of 3,4-dihydro-2H-1,4-benzoxazine, beta-adrenergic affinity, inotropic, chronotropic and coronary vasodilating activities. Eur. J. Med. Chem.
**1999**, 34, 903–917. [Google Scholar] [CrossRef] - DaCambra, M.P.; Yusta, B.; Sumner-Smith, M.; Crivici, A.; Drucker, D.J.; Brubaker, P.L. Structural determinants for activity of glucagon-like peptide-2. Biochemistry
**2000**, 39, 8888–8894. [Google Scholar] [CrossRef] [PubMed] - Mukai, Y.; Shibata, H.; Nakamura, T.; Yoshioka, Y.; Abe, Y.; Nomura, T.; Taniai, M.; Ohta, T.; Ikemizu, S.; Nakagawa, S.; et al. Structure–Function Relationship of Tumor Necrosis Factor (TNF) and Its Receptor Interaction Based on 3D Structural Analysis of a Fully Active TNFR1-Selective TNF Mutant. J. Mol. Biol.
**2009**, 385, 1221–1229. [Google Scholar] [CrossRef] [PubMed]

**Figure 1.**Encoding of sequences with innov’SAR: (

**a**) One index encoding: only one index is used and leads to an elementary sequence Ele_(SEQ), and (

**b**) Example of an extended sequence of three indexes with FFT: FFT_Seq

_{ja}—FFT_Seq

_{jb}—FFT_Seq

_{jc}: three indices are concatenated and lead to extended numerical sequence (Ext_SEQ). (

**c**) A general view of the different phases of innov’SAR with extended sequences. An encoding phase transforms the primary sequences of the initial dataset into protein spectra. The spectra are concatenated to generate extended sequences. The modeling phase uses the extended sequences and the protein activity as a learning dataset in order to construct a regression model. Next, the model performances are evaluated by a comparison between measured and predicted activities.

**Figure 2.**Plot of measured thermostability of cytochrome P450 variants versus predicted thermostability using innov’SAR algorithm with (

**a**) one index coupled with Fast Fourier Transform (FFT), (

**b**) with the same index coupled and not coupled with FFT and, (

**c**) two indices, one coupled with FFT and the other not. The number of the index in the amino acid index (AAindex) database is indicated at the top of the plot: “300”.

**Figure 3.**Plot of measured potency (fold-increase in cyclic adenosine monophosphate, cAMP, concentration) of Glucagon-like peptide-2 (GLP-2) variants versus predicted potency using innov’SAR algorithm with one index coupled with Fast Fourier Transform (FFT).

**Figure 4.**Plot of measured ΔΔG

^{‡}of epoxide hydrolase variants versus predicted ΔΔG

^{‡}using innov’SAR algorithm with one index coupled with Fast Fourier Transform (FFT).

**Figure 5.**Workflow of the iterative process for successive concatenation. Each round uses the indices from the previous iteration as a base of extended sequence (Ext_SEQ) and determines the best index to keep for the current round by evaluation of the modeling performances.

**Figure 6.**Plot of measured potency (fold-increase in cyclic adenosine monophosphate, cAMP, concentration) of Glucagon-like peptide-2 (GLP-2) variants versus predicted potency using innov’SAR algorithm with three indices coupled with Fast Fourier Transform (FFT).

**Figure 7.**Plot of measured ΔΔG

^{‡}of epoxide hydrolase variants versus predicted ΔΔG

^{‡}using innov’SAR algorithm with three indices coupled with Fast Fourier Transform (FFT).

**Figure 8.**Plot of measured affinity versus predicted affinity of tumor necrosis factor (TNF) variants using (

**a**) a single index or (

**b**) three indices coupled with Fast Fourier Transform (FFT).

**Figure 9.**Plot of measured versus predicted thermostability of cytochrome P450 variants using three indices coupled with Fast Fourier Transform (FFT).

**Figure 10.**Plot of measured versus predicted thermostability of cytochrome P450 variants using (

**a**) a single index (300) coupled with Fast Fourier Transform (FFT) and a selection of 20% frequencies, (

**b**) a single index (300) twice, once without FFT and another time coupled with FFT and by selection of 20% frequencies, (

**c**) using two indices, one index without FFT (343) and one index coupled with FFT (300) and by selection of 20% frequencies, and (

**d**) using innov’SAR algorithm with three indices (300, 39, 226) coupled with FFT and by selection of 20% frequencies.

**Table 1.**GLP-2: Top 10 extended sequences selected using a combinatorial approach applied on the top 10 best indices.

Index | cvRMSE | cvR^{2} |
---|---|---|

440 350 44 | 1.99 | 0.47 |

440 350 | 1.99 | 0.47 |

440 350 233 | 1.99 | 0.47 |

440 44 | 2.06 | 0.37 |

44 233 | 2.09 | 0.37 |

449 350 | 2.10 | 0.43 |

449 350 233 | 2.10 | 0.43 |

449 350 44 | 2.10 | 0.43 |

449 | 2.11 | 0.42 |

449 233 | 2.11 | 0.42 |

**Table 2.**Epoxide hydrolase: Top 10 extended sequences selected using a combinatorial applied to the top 10 best indices.

Index | cvRMSE | cvR^{2} |
---|---|---|

161 178 516 | 0.1051 | 0.9685 |

254 178 516 | 0.1051 | 0.9685 |

232 161 508 | 0.1123 | 0.9640 |

232 254 508 | 0.1123 | 0.9640 |

161 508 | 0.1146 | 0.9629 |

254 508 | 0.1146 | 0.9629 |

161 254 508 | 0.1150 | 0.9624 |

303 508 | 0.1161 | 0.9624 |

303 161 508 | 0.1170 | 0.9615 |

303 254 508 | 0.1170 | 0.9615 |

Dataset | Index 1 and Protein Feature | Index 2 and Protein Feature | Index 3 and Protein Feature |
---|---|---|---|

GLP-2 | Index 449, other properties family | Index 341, alpha and turn propensities family | Index 193, composition family |

Epoxide hydrolase | Index 303, other properties family | Index 14, hydrophobicity | Index 234, beta propensity |

TNF alpha | Index 203, composition | Index 504, NA | Index 486, NA |

Cytochrome P450 | Index 300, alpha and turn propensities | Index 39, beta propensity | Index 226 beta propensity |

**Table 4.**Summary of the performance metrics values obtained on the four datasets and for the different kinds of experiments carried out. “-“ indicates that the corresponding model is used as a reference for calculation of the p-value.

Dataset | cvRMSE | cvR^{2} | p-Value (Error² with Paired Student Test) |
---|---|---|---|

Cytochrome P450 | |||

Cytochrome P450 FFT one index Figure 2a | 1.91 | 0.83 | - |

Cytochrome P450 multi indices noFFT_FFT Figure 2c | 1.85 | 0.84 | 0.4123 |

Cytochrome P450 one index noFFT + FFT Figure 2b | 1.89 | 0.83 | 0.7781 |

Cytochrome P450 3 indices FFT Figure 9 | 1.63 | 0.88 | 0.0020 |

Cytochrome P450 - selection of 20% frequencies | |||

Cytochrome P450 1 index FFT selection 20% Figure 10a | 2.68 | 0.66 | - |

Cytochrome P450 two indices noFFT_FFT selection 20% Figure 10c | 2.39 | 0.74 | 0.0767 |

Cytochrome P450 one index noFFT + FFT selection 20% (Figure 10b) | 2.38 | 0.74 | 0.0600 |

Cytochrome P450 3 indices FFT selection 20% Figure 10d | 2.52 | 0.7 | 0.0826 |

GLP-2 | |||

GLP-2 1 index FFT Figure 3 | 2.11 | 0.42 | - |

GLP-2 3 indices FFT Figure 6 | 1.75 | 0.55 | 0.0078 |

GLP-2 3 indices best in Table 1 | 1.99 | 0.47 | 0.5309141 |

Epoxide Hydrolase | |||

Epoxide Hydrolase 1 index FFT 303 Figure 4 | 0.12 | 0.96 | - |

Epoxide Hydrolase best in Table 2 | 0.1051 | 0.9685 | 0.4322954 |

Epoxide Hydrolase 3 indices FFT Figure 7 | 0.094 | 0.9747 | 0.3435683 |

TNF Alpha | |||

TNF 1 index 203 Figure 8a | 0.32 | 0.85 | - |

TNF 3 indices FFT Figure 8b | 0.28 | 0.88 | 0.1749 |

**Table 5.**Correspondence of the index number and the name and reference of the index in the AAindex database [27].

Index Number | Index Name | Applied on Dataset |
---|---|---|

39 | Normalized frequency of beta-sheet [30] | CYP450 |

226 | Normalized frequency of beta-sheet from CF [31] | CYP450 |

300 | Average relative fractional occurrence in A0(i) [32] | CYP450 |

343 | Information measure for extended [33] | CYP450 |

450 | Hydropathy scale based on self-information values in the two-state model (25% accessibility) [34] | CYP450 |

14 | Transfer free energy to surface [35] | Epoxide H. |

161 | Normalized frequency of beta-sheet, with weights [36] | Epoxide H. |

178 | Retention coefficient in HPLC, pH7.4 [37] | Epoxide H. |

232 | Normalized frequency of beta-sheet in all-beta class [31] | Epoxide H. |

254 | Relative frequency in beta-sheet [38] | Epoxide H. |

303 | Average relative fractional occurrence in EL(i) [32] | Epoxide H. |

508 | Linker propensity from helical (annotated by DSSP) dataset [39] | Epoxide H. |

516 | Hydrostatic pressure asymmetry index, PAI [40] | Epoxide H. |

44 | Normalized frequency of C-terminal non-helical region [30] | GLP-2 |

193 | AA composition of mt-proteins from animal [41] | GLP-2 |

233 | Normalized frequency of beta-sheet in alpha + beta class [39] | GLP-2 |

341 | Information measure for middle helix [33] | GLP-2 |

350 | Information measure for coil [33] | GLP-2 |

440 | Distribution of amino acid residues in the 18 non-redundant families of thermophilic proteins [42] | GLP-2 |

449 | Hydropathy scale based on self-information values in the two-state model (20% accessibility) [34] | GLP-2 |

203 | AA composition of CYT2 of single-spanning proteins [43] | TNF |

297 | Average reduced distance for C-alpha [44] | TNF |

486 | Electron-ion interaction potential values [17] | TNF |

504 | Linker propensity from three-linker dataset [39] | TNF |

523 | Apparent partition energies calculated from Chothia index [45] | TNF |

Dataset | Calculation Time for Method A (min) | Calculation Time for Method B (min) |
---|---|---|

GLP-2 | 5 | 2 |

Epoxide hydrolase | 60 | 9 |

TNF alpha | 15 | 3 |

Cytochrome P450 | 120 | ND |

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Fontaine, N.T.; Cadet, X.F.; Vetrivel, I.
Novel Descriptors and Digital Signal Processing- Based Method for Protein Sequence Activity Relationship Study. *Int. J. Mol. Sci.* **2019**, *20*, 5640.
https://doi.org/10.3390/ijms20225640

**AMA Style**

Fontaine NT, Cadet XF, Vetrivel I.
Novel Descriptors and Digital Signal Processing- Based Method for Protein Sequence Activity Relationship Study. *International Journal of Molecular Sciences*. 2019; 20(22):5640.
https://doi.org/10.3390/ijms20225640

**Chicago/Turabian Style**

Fontaine, Nicolas T., Xavier F. Cadet, and Iyanar Vetrivel.
2019. "Novel Descriptors and Digital Signal Processing- Based Method for Protein Sequence Activity Relationship Study" *International Journal of Molecular Sciences* 20, no. 22: 5640.
https://doi.org/10.3390/ijms20225640