Improved prediction and characterization of anticancer activities of peptides using a novel flexible scoring card method

Abstract

As anticancer peptides (ACPs) have attracted great interest for cancer treatment, several approaches based on machine learning have been proposed for ACP identification. Although existing methods have afforded high prediction accuracies, however such models are using a large number of descriptors together with complex ensemble approaches that consequently leads to low interpretability and thus poses a challenge for biologists and biochemists. Therefore, it is desirable to develop a simple, interpretable and efficient predictor for accurate ACP identification as well as providing the means for the rational design of new anticancer peptides with promising potential for clinical application. Herein, we propose a novel flexible scoring card method (FSCM) making use of propensity scores of local and global sequential information for the development of a sequence-based ACP predictor (named iACP-FSCM) for improving the prediction accuracy and model interpretability. To the best of our knowledge, iACP-FSCM represents the first sequence-based ACP predictor for rationalizing an in-depth understanding into the molecular basis for the enhancement of anticancer activities of peptides via the use of FSCM-derived propensity scores. The independent testing results showed that the iACP-FSCM provided accuracies of 0.825 and 0.910 as evaluated on the main and alternative datasets, respectively. Results from comparative benchmarking demonstrated that iACP-FSCM could outperform seven other existing ACP predictors with marked improvements of 7% and 17% for accuracy and MCC, respectively, on the main dataset. Furthermore, the iACP-FSCM (0.910) achieved very comparable results to that of the state-of-the-art ensemble model AntiCP2.0 (0.920) as evaluated on the alternative dataset. Comparative results demonstrated that iACP-FSCM was the most suitable choice for ACP identification and characterization considering its simplicity, interpretability and generalizability. It is highly anticipated that the iACP-FSCM may be a robust tool for the rapid screening and identification of promising ACPs for clinical use.

ACP-MHCNN: an accurate multi-headed deep-convolutional neural network to predict anticancer peptides

“…We also compare ACP-MHCNN with several existing ACP identification methods on both main and alternate datasets used in 17 , and the results are shown in Table 12 . This comparison shows that ACPred-LAF 16 , iACP-FSCM 57 , and AntiCP-2.0 17 slightly outperforms ACP-MHCNN, and all outperform other existing methods by significant margin on these two specific datasets. It is worth noting that, since AntiCP-2.0 and all of the existing methods reported in Table 12 are traditional machine learning models while ACP-MHCNN is composed of several convolutional layers with much larger effective hypotheses space, the sizes of the training partitions of main and alternate datasets are the bottleneck for ACP-MHCNN when it comes to generalization capability.…”

“…During the revision stage of this manuscript, Charoenkwan et al proposed a sequence-based method iACP-FSCM with an emphasis on model interpretability, where 11 local and global amino acid composition-based features were utilized with a weighted-sum-based prediction mechanism 16 . Furthermore, Agrawal et al proposed a sequence-based method AntiCP 2.0 along with two ACP identification datasets 17 .…”

A novel sequence-based predictor for identifying and characterizing thermophilic proteins using estimated propensity scores of dipeptides

“…The SCM method has been demonstrated to perform admirably in terms of conceptual simplicity, ease of implementation and interpretability 16 , 18 , 36 – 39 . In 2012, Huang et al 19 firstly introduced the original SCM method.…”

Applications of Virtual Screening in Bioprospecting: Facts, Shifts, and Perspectives to Explore the Chemo-Structural Diversity of Natural Products

“…Artificial Intelligence has made important progress toward the acceleration of research and development of novel bioactive natural compounds with industrial applications. This approach has been widely applied in different steps related to the virtual screening strategies, for example to predict some pharmacokinetic properties (Wei et al, 2017 ; Qiang et al, 2018 ) [e.g., penetration of compounds into the blood–brain barrier (Zhang et al, 2017 ; Dai et al, 2021 ) and cell membrane (Wei et al, 2017 ; Wolfe et al, 2018 )], compounds' side effects (Dimitri and Lió, 2017 ), their toxicity (Mayr et al, 2016 ; Pu et al, 2019 ; Zheng et al, 2020 ), molecular targets (Wang et al, 2013 ; Jeon et al, 2014 ), and their bioactivity (Li and Huang, 2012 ; Schaduangrat et al, 2019 ; Shoombuatong et al, 2019 ) [e.g., anti-tuberculosis (Gomes et al, 2017 ; Maia S. M. et al, 2020 ), anticancer (Charoenkwan et al, 2021 ), and insecticidal activities (Soares Rodrigues et al, 2021 )] as well as to identify the pan-assay interference compounds (PAINS), i.e., highly reactive and promiscuous molecules that are often false positives in high-throughput screening assays (Jasial et al, 2018 ). In some cases, the ML algorithms have been reported with superior efficiency and, thus, are more suitable to predict hit compounds from chemical libraries than are the traditional QSAR methods (Tsou et al, 2020 ).…”