Combination of digital signal processing and assembled predictive models facilitates the rational design of proteins

Abstract

Predicting the effect of mutations in proteins is one of the most critical challenges in protein engineering; by knowing the effect a substitution of one (or several) residues in the protein's sequence has on its overall properties, could design a variant with a desirable function. New strategies and methodologies to create predictive models are continually being developed. However, those that claim to be general often do not reach adequate performance, and those that aim to a particular task improve their predictive performance at the cost of the method's generality. Moreover, these approaches typically require a particular decision to encode the amino acidic sequence, without an explicit methodological agreement in such endeavor. To address these issues, in this work, we applied clustering, embedding, and dimensionality reduction techniques to the AAIndex database to select meaningful combinations of physicochemical properties for the encoding stage. We then used the chosen set of properties to obtain several encodings of the same sequence, to subsequently apply the Fast Fourier Transform (FFT) on them. We perform an exploratory stage of Machine-Learning models in the frequency space, using different algorithms and hyperparameters. Finally, we select the best performing predictive models in each set of properties and create an assembled model. We extensively tested the proposed methodology on different datasets and demonstrated that the generated assembled model achieved notably better performance metrics than those models based on a single encoding and, in most cases, better than those previously reported. The proposed method is available as a Python library for non-commercial use under the GNU General Public License (GPLv3) license.K eywords Protein Engineering -predictive models • Protein Engineering -rational design • machinelearning algorithms • digital signal processing • assembled models

Peptipedia: a comprehensive database for peptide research supported by Assembled predictive models and Data Mining approaches

“…Remarkably, all the models generated presented an accuracy of over 83% (see Table 1 and section 5 of the Support Information for details). We previously compared the results obtained by applying this type of strategies against classical sequence coding methods, demonstrating better results (Medina-Ortiz et al, 2020a). Furthermore, we compare our results with previously developed classification models for peptide sequences.…”

“…One of the essential services of Peptipedia is the activity classification system for peptide sequences based on Machine Learning strategies. The training of models was based on the application of supervised learning algorithms combined with sequence coding approaches, using physicochemical properties and Digital Signal Processing, according to the strategies proposed by Medina-Ortiz et al (2020a). In this way, we generated assembled binary models to recognize activities for peptide sequences employing our categories proposed in this work.…”

“…Using coding of physicochemical properties and their representation in frequency space (Medina-Ortiz et al, 2020a) and employing recursive binary division strategies to optimise performance measures (Medina-Ortiz et al, 2020c) we depeloped 44 assembled binary models for classification of activity for peptide sequences, considering the categories and subcategories proposed in this work. We used k-fold cross-validation to avoid model overfitting.…”

“…Even though we present a lower performance in particular situations than previously developed methods. Nevertheless, the proposed strategy is generic, could be apply in activity classification of peptides sequences problems, prediction of properties, and multiple issues in protein engineering (Medina-Ortiz et al, 2020a). Notably, we validated all our models using statistical methods.…”

Peptipedia v2.0: A peptide sequence database and user-friendly web platform. A major update

“…A K-fold cross-validation (k = 10) was performed to prevent overfitting. The validation dataset was then used to assess model performance using classical metrics such as precision, recall, accuracy, and F-score (Medina-Ortiz et al, 2020a).…”