Background:It seems to be numerous unclear black-box mechanisms of Chinese Medicines(CMs)with multiple bioactivities in the real-world clinical practice.Meanwhile,prior prediction is necessary before the implementatio...Background:It seems to be numerous unclear black-box mechanisms of Chinese Medicines(CMs)with multiple bioactivities in the real-world clinical practice.Meanwhile,prior prediction is necessary before the implementation of pharmacodynamics-pharmacokinetics-based researches.With emergent ML techniques for TCM domain,Bayesian Network(BN)has shown its potentials for CM-bioactivity prediction and syndromes identification in Traditional Chinese Medicine(TCM),benefited from many advantages,such as flexibility in addressing,data-driven and probability-based inference under complex uncertainty.Although BN has been extensively used in TCM,the scarcity of researches on refining methodological features of BN-modelling for optimization poses a significant challenge.Our goal is to present methodological overview of BN-modelling for CM-bioactivities prediction towards pharmacology,which tends to acquire a sequence of intimations for boosting in-depth and optimized CM-BN collaboration based on detected gaps.Methods:We performed systematic search of 13 databases from their inception to November 10th 2022 regardless of language written,which excluded unindexed journals and clinical trial registries,using the 3 keywords(CM,Pharmacology,BN).And full-text original researches with the given subject were under consideration.Afterwards,selection of eligible studies,data refinement and inspection were totally conducted by 6 review authors.Results:A total of 7 studies involving 17 BN models were included for synthesis and refinement,based on existing literatures and databases with 2 modelling functions:regression and tagging.There were 3 prediction patterns:property-bioactivity,efficacy-bioactivity and constituent-bioactivity inference,covering 8 feature-utilized efficacies,5 feature-utilized properties and 10 feature-utilized constituents.Thereafter,without an independent validation dataset,established BNs were mostly utilized to predict the root-node probabilities of unknown data.Indeed,incomplete report on modelling samples,directed acyclic graphs,conditional probability tables and algorithms hindered us from gathering information.Conclusion:A spot of studies were found in this work.And current evidence suggested that some breakthroughs should be achieved in CM-BN integration in the future.At last,to our knowledge,we preliminarily proposed certain recommendations and elicited implications for future work.展开更多
基金supported by National Natural Science Foundation of China(No.82104943)。
文摘Background:It seems to be numerous unclear black-box mechanisms of Chinese Medicines(CMs)with multiple bioactivities in the real-world clinical practice.Meanwhile,prior prediction is necessary before the implementation of pharmacodynamics-pharmacokinetics-based researches.With emergent ML techniques for TCM domain,Bayesian Network(BN)has shown its potentials for CM-bioactivity prediction and syndromes identification in Traditional Chinese Medicine(TCM),benefited from many advantages,such as flexibility in addressing,data-driven and probability-based inference under complex uncertainty.Although BN has been extensively used in TCM,the scarcity of researches on refining methodological features of BN-modelling for optimization poses a significant challenge.Our goal is to present methodological overview of BN-modelling for CM-bioactivities prediction towards pharmacology,which tends to acquire a sequence of intimations for boosting in-depth and optimized CM-BN collaboration based on detected gaps.Methods:We performed systematic search of 13 databases from their inception to November 10th 2022 regardless of language written,which excluded unindexed journals and clinical trial registries,using the 3 keywords(CM,Pharmacology,BN).And full-text original researches with the given subject were under consideration.Afterwards,selection of eligible studies,data refinement and inspection were totally conducted by 6 review authors.Results:A total of 7 studies involving 17 BN models were included for synthesis and refinement,based on existing literatures and databases with 2 modelling functions:regression and tagging.There were 3 prediction patterns:property-bioactivity,efficacy-bioactivity and constituent-bioactivity inference,covering 8 feature-utilized efficacies,5 feature-utilized properties and 10 feature-utilized constituents.Thereafter,without an independent validation dataset,established BNs were mostly utilized to predict the root-node probabilities of unknown data.Indeed,incomplete report on modelling samples,directed acyclic graphs,conditional probability tables and algorithms hindered us from gathering information.Conclusion:A spot of studies were found in this work.And current evidence suggested that some breakthroughs should be achieved in CM-BN integration in the future.At last,to our knowledge,we preliminarily proposed certain recommendations and elicited implications for future work.
