Exploring local discriminative information from evolutionary profiles for cytokine–receptor interaction prediction

Wei, Leyi; Zhang, Bowen; Chen, Zhiyong; Gao, Xing; Liao, Minghong

doi:10.1016/j.neucom.2016.02.078

Cited by 13 publications

(13 citation statements)

References 74 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using this feature set, we obtained a SMO-RBF model with 90.1% accuracy, an mcc value of 0.805, and a g-means value of 90.1—the highest values over all algorithms and feature configurations (see Table 1 ). Moreover, the model was 2.2% more accurate than the model previously described by Wei et al [ 22 ], which had been trained on the same dataset. The highest AUC value among all trained models with this particular feature set was obtained with the RF model (0.935).…”

Section: Resultsmentioning

confidence: 65%

“…The trained ML models were thoroughly evaluated using the popular leave one out cross validation, 10 × 10-fold cross validation, and 70% training/30% testing split. Our results show, performances of all ML algorithms benefitted from K -means based sampling, with RF models performing best, significantly outperforming the Wei et al [ 22 ] model with respect to all metrics except sensitivity. Negative dataset selection as well as sampling have significant impact on prediction performance.…”

Section: Discussionmentioning

confidence: 75%

“…Each of the selected ML algorithms was applied using the original dataset by Wei et al [ 22 ] and various feature sets (see “ Methods ” below and Additional file 1 ). The performance was evaluated based on loocv (Additional file 1 : Tables S1–S12 and Table 1 ).…”

Section: Resultsmentioning

confidence: 99%

“…Of note, we achieved a higher accuracy with a smaller feature set, which has the additional benefit of a shorter classifier training time. Our feature vector had only 112 (2 × 56) dimensions as compared to 1600 dimensions of the feature vectors used in [ 22 ].…”

Section: Resultsmentioning

confidence: 99%

“…Since the specificity is higher than the sensitivity, it can be concluded that their prediction model is more accurate for predicting non-interacting pairs than interacting pairs. Wei et al [ 22 ] further improved the CRI prediction accuracy using random forest with evolutionary features, i.e. Pseudo Position-Specific Score Matrix (Pse-PSSM) and amino acid composition and PSSM with auto-covariance (AC) transformation (AAC_PSSM_AC).…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Improved cytokine–receptor interaction prediction by exploiting the negative sample space

Nath

Leier

2020

BMC Bioinformatics

View full text Add to dashboard Cite

Background Cytokines act by binding to specific receptors in the plasma membrane of target cells. Knowledge of cytokine–receptor interaction (CRI) is very important for understanding the pathogenesis of various human diseases—notably autoimmune, inflammatory and infectious diseases—and identifying potential therapeutic targets. Recently, machine learning algorithms have been used to predict CRIs. “Gold Standard” negative datasets are still lacking and strong biases in negative datasets can significantly affect the training of learning algorithms and their evaluation. To mitigate the unrepresentativeness and bias inherent in the negative sample selection (non-interacting proteins), we propose a clustering-based approach for representative negative sample selection. Results We used deep autoencoders to investigate the effect of different sampling approaches for non-interacting pairs on the training and the performance of machine learning classifiers. By using the anomaly detection capabilities of deep autoencoders we deduced the effects of different categories of negative samples on the training of learning algorithms. Random sampling for selecting non-interacting pairs results in either over- or under-representation of hard or easy to classify instances. When K-means based sampling of negative datasets is applied to mitigate the inadequacies of random sampling, random forest (RF) together with the combined feature set of atomic composition, physicochemical-2grams and two different representations of evolutionary information performs best. Average model performances based on leave-one-out cross validation (loocv) over ten different negative sample sets that each model was trained with, show that RF models significantly outperform the previous best CRI predictor in terms of accuracy (+ 5.1%), specificity (+ 13%), mcc (+ 0.1) and g-means value (+ 5.1). Evaluations using tenfold cv and training/testing splits confirm the competitive performance. Conclusions A comparative analysis was performed to assess the effect of three different sampling methods (random, K-means and uniform sampling) on the training of learning algorithms using different evaluation methods. Models trained on K-means sampled datasets generally show a significantly improved performance compared to those trained on random selections—with RF seemingly benefiting most in our particular setting. Our findings on the sampling are highly relevant and apply to many applications of supervised learning approaches in bioinformatics.

show abstract

Section: Resultsmentioning

confidence: 65%

Section: Discussionmentioning

confidence: 75%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations