Boosting phosphorylation site prediction with sequence feature‐based machine learning

Abstract

Protein phosphorylation is one of the essential posttranslation modifications playing a vital role in the regulation of many fundamental cellular processes. We propose a LightGBM‐based computational approach that uses evolutionary, geometric, sequence environment, and amino acid‐specific features to decipher phosphate binding sites from a protein sequence. Our method, while compared with other existing methods on 2429 protein sequences taken from standard Phospho.ELM (P.ELM) benchmark data set featuring 11 organisms reports a higher F1 score = 0.504 (harmonic mean of the precision and recall) and ROC AUC = 0.836 (area under the curve of the receiver operating characteristics). The computation time of our proposed approach is much less than that of the recently developed deep learning‐based framework. Structural analysis on selected protein sequences informs that our prediction is the superset of the phosphorylation sites, as mentioned in P.ELM data set. The foundation of our scheme is manual feature engineering and a decision tree‐based classification. Hence, it is intuitive, and one can interpret the final tree as a set of rules resulting in a deeper understanding of the relationships between biophysical features and phosphorylation sites. Our innovative problem transformation method permits more control over precision and recall as is demonstrated by the fact that if we incorporate output probability of the existing deep learning framework as an additional feature, then our prediction improves (F1 score = 0.546; ROC AUC = 0.849). The implementation of our method can be accessed at http://cse.iitkgp.ac.in/~pralay/resources/PPSBoost/ and is mirrored at https://cosmos.iitkgp.ac.in/PPSBoost.

Predicting phosphorylation sites using machine learning by integrating the sequence, structure, and functional information of proteins

“…With regard to post-mRMR and SU attribute selection, of the PP-based features, AAC, molecular weight, residue volume, flexibility, and partition coefficient of predominantly AA11 turned out to be highly significant for classification, followed by the hydrophobicity of the AAs around the central residue. These features have already been shown to be relevant for discriminating between phosphorylated and non-phosphorylated sites [ 13 , 18 ]. Table 2 lists the initial number of different features types used to encode sequence fragments.…”

“…Although these techniques provide a vast amount of data when operated in a high-throughput manner, they are laborious, costly, time-consuming, and often produce false positives and false negatives. A large number of PTMs thus remain unidentified or misclassified, and the associated mechanisms in context of cellular and biological processes are overlooked [ 13 ]. The computational prediction of protein phosphorylation sites appears to be a promising alternative strategy for reducing the associated costs and time.…”

“…Around 40 computational methods have been developed to predict protein phosphorylation sites [ 14 ]. For example, Maiti et al proposed an approach that uses sequence-environment-specific, geometric, and evolutionary information-based features to identify phosphorylation sites using the LightGBM algorithm [ 13 ]. Other machine-learning (ML)-based approaches, such as PhosPred-RF [ 11 ] and PhosphoSVM [ 15 ], use only sequence-based features for predictions based on random forest (RF) and support vector machine (SVM), respectively.…”

Staem5: A novel computational approach for accurate prediction of m5C site

“…26 In addition, LightGBM proposes gradient-based one-side sampling, exclusive feature bundling, and leaf-wise growth strategy to obtain better accuracy and efficient computation. Meanwhile, it also adopted limiting maximum depth parameters to mitigate over-fitting 55 , 56 and LightGBM has been widely used in bioinformatics. 57 , 58 ExtraTree is also a tree-based algorithm that was proposed by Pierre Geurts et al.…”

Machine Learning-Based Application for Predicting Risk of Type 2 Diabetes Mellitus (T2DM) in Saudi Arabia: A Retrospective Cross-Sectional Study

“…In our case, both precision and recall are essential since we were interested in finding the association of the target variable (high risk of diabetes) with the explanatory variable(s). So, we selected a classification algorithm that maximizes an F1 score metric, a harmonic mean of both recall and precision [47]. A set of classification algorithms was screened based on the F1 score, and the algorithm with the highest F1 score was selected.…”