AntAngioCOOL: computational detection of anti-angiogenic peptides

Abstract

Background Angiogenesis inhibition research is a cutting edge area in angiogenesis-dependent disease therapy, especially in cancer therapy. Recently, studies on anti-angiogenic peptides have provided promising results in the field of cancer treatment. Methods A non-redundant dataset of 135 anti-angiogenic peptides (positive instances) and 135 non anti-angiogenic peptides (negative instances) was used in this study. Also, 20% of each class were selected to construct an independent test dataset (see Additional files 1, 2). We proposed an effective machine learning based R package (AntAngioCOOL) to predict anti-angiogenic peptides. We have examined more than 200 different classifiers to build an efficient predictor. Also, more than 17,000 features were extracted to encode the peptides. Results Finally, more than 2000 informative features were selected to train the classifiers for detecting anti-angiogenic peptides. AntAngioCOOL includes three different models that can be selected by the user for different purposes; it is the most sensitive, most specific and most accurate. According to the obtained results AntAngioCOOL can effectively suggest anti-angiogenic peptides; this tool achieved sensitivity of 88%, specificity of 77% and accuracy of 75% on the independent test set. AntAngioCOOL can be accessed at https://cran.r-project.org/ . Conclusions Only 2% of the extracted descriptors were used to build the predictor models. The results revealed that physico-chemical profile is the most important feature type in predicting anti-angiogenic peptides. Also, atomic profile and PseAAC are the other important features. Electronic supplementary material The online version of this article (10.1186/s12967-019-1813-7) contains supplementary material, which is available to authorized users.

TargetAntiAngio: A Sequence-Based Tool for the Prediction and Analysis of Anti-Angiogenic Peptides

“…The first and most important consideration for developing a promising computational model is to construct a reliable benchmark dataset. In this study, the benchmark dataset was obtained from the work of Ramaprasad et al [27], which has been used for developing recent prediction models of anti-angiogenic peptides [28,29]. Initially, the benchmark dataset had 257 peptide sequences in the anti-angiogenic class as derived from various articles and patents.…”

“…Their comparison results indicated that the best model based on generalized linear model yielded 86% accuracy in which such model utilized the top 200 informative features for model building. Recently, Zahiri et al [29] developed the AntAngioCOOL R package and executed performance comparisons amongst various machine learning techniques and types of peptide features. Based on their results on performance comparisons, regression, and survival trees model employing descriptors consisting of pseudo amino acid composition, k -mer composition, k -mer composition (reduced alphabet), physico-chemical profile, and atomic profile yielded the highest accuracy of 77% over an independent validation test from 1 round of random split.…”

“…Although all these methods yielded encouraging results and played an important role in simulating the development of anti-angiogenic peptides identification, there is still room for further improvement in the prediction performance of anti-angiogenic peptides. The following research gaps have been elucidated as follows: (i) Blanco et al’s method [28] and AntAngioCOOL [29] were assessed by the independent validation test from only 1 round of random split, hence, their prediction results were not yet satisfactory; (ii) the studies of Blanco et al’s method [28] and AntAngioCOOL [29] did not provide a web server, hence, their usage was quite limited; and (iii) AntiAngioPred [27] and Blanco et al’s method were not straight-forward enough to provide the underlying mechanism of anti-angiogenic peptides due of the lack of interpretability of the model.…”

SARS-CoV-2-human protein-protein interaction network

“…Random forest was the approach followed by both Yang [ 25 ] and Barman [ 26 ]. The approach followed by Leite [ 27 ], Zahiri [ 28 ], and Mei [ 29 ], were k-nearest neighbors, multilayer perceptron, and AdaBoost, respectively.…”

“…On the other hand, researchers work on different viruses to predict their PPI networks. Zhao predicted HIV1-human PPI network [ 28 ], Khorsand uncovered Alphainfluenzavirus-human PPI network [ 29 ], Ray predicted HCV-human PPI network [ 30 ], Chan revealed west nile virus-human PPI network [ 31 ], and Duran worked on herpesvirus-human PPI network [ 32 ].…”