The Research On Biclustering Algorithm And Its Application In Chinese Adverse Drug Reaction Monitoring

Background and aims:As an indispensable part of safety surveillance of post-marketing drugs,Spontaneous reporting system(SRS)can collect all kinds of adverse drug reactions(ADR).After analyzing the data in spontaneous reporting system,researchers can effectively monitor the state of adverse drug reactions.There was a large amount of ADR reports in Chinese SRS.The number of reports surged above 1.4 million in 2016,while this number was just 0.60 million in 2009.Using classical disproportionate measures,lots of signals can be found out and can provide evidences for drugs safety evaluation.However,there wiil be lots of work needed to be done before a signal was finally identified as a real ADR.The vast amount of signals presents a challenge to evaluation.At present,the main ways of signal evaluation include contrasting with known ADR,expert evaluation and this work is a time-consuming and low efficient process.How to enhance the efficiency of signal evaluation is one of the issues that needed to be addressed as a matter of urgency.A previous research pointed out that a single drug-ADR combination should not be considered in isolation,but combinations of other ADRs with that drug and other drugs with that same ADR should be considered.This provides new ideas for signal evaluation work.This study aims to use a new analysis method: biclustering analysis to find out similar signals groups.Using known ADEs in groups,researchers can evaluate unrecognized ADEs.Through estimating best paramets when using biclustering analysis,this study will finally increase the efficiency of signal evaluation.Methods:This study use the 2010-2011 data in China SRS as data source.The ADR signals were transformed into original quantitative data matrix.Based on different strength thresholds of IC,the original quantitative data matrix will be transformed into corresponding 0-1 data matrix.Using different Bimax algorithm parameters,we can finally obtaine biclustering results and find out combinations of ADRs and drugs of which all of the assocations are “strength signals”.Based on the rank sum ration(RSR)comprehensive evaluation and using the average coincidence ratio,involving ratio,this study finally determined the most suitable parameters.Then our study will examine the biclusting result under the most suitable parameters and evaluation the effect of biclustering analysis in Chinese signals data.Result:Through RSR comprehensive evaluation,the most suitable parameters were identified as IC strength threshold: 0.80,Bimax algorithm parameters: 3 as the minimum for both rows and columnsUnder these parameters.biclustering analysis can recognize 4293 biclusters signals data.The average number of signals in biclusters was 23.41.The average number of IC value in biclusters was 2.38.Through taxonomic evaluation,our study found that 1835 biclusters(42.8%)contained similar drugs,simailr ADRs,which contained 72.3% of signals whose IC was no less than 0.80.4272 biclusters(99.5%)contained at least one signal which has been comfirmed.All of the signals contained in 193 biclusters(4.5%)have been comfirmed.The most common drug in biclusters was “Risperidone”(included in 708 biclusters,16.5%)and the most common ADR was “abnormal liver function”(included in 720 biclusters,16.8%).Conclusions:The biclusters recognized by biclustering analysis can provide support for confirming unrecognized ADRs,screening drug-ADR combinations which should be paid more attention and so on.Biclustering analysis can increase the efficiency of signals evaluation in China.