'Bingo' - a large language model- and graph neural network (LLM-GNN)-based workflow for the prediction of essential genes from protein data

发布时间：2024-07-20 点击次数：

• 影响因子：9.5
• 发表刊物：Briefings in Bioinformatics
• 关键字：essential gene prediction; large language model; graph neural network; adversarial training; biological interpretation
• 摘要：Theidentificationandcharacterizationofessentialgenesarecentraltoourunderstandingofthecorebiologicalfunctionsineukaryoticorganisms,andhasimportantimplicationsforthetreatmentofdiseasescausedby,forexample,cancersandpathogens.Giventhemajorconstraintsintestingthefunctionsofgenesofmanyorganismsinthelaboratory,duetotheabsenceofinvitroculturesand/orgeneperturbationassaysformostmetazoanspecies,therehasbeenaneedtodevelopinsilicotoolsfortheaccuratepredictionorinferenceofessentialgenestounderpinsystemsbiologicalinvestigations.Majoradvancesinmachinelearningapproachesprovideunprecedentedopportunitiestoovercometheselimitationsandacceleratethediscoveryofessentialgenesonagenome-widescale.Here,wedevelopedandevaluatedalargelanguagemodel-andgraphneuralnetwork(LLM–GNN)-basedapproach,called‘Bingo’,topredictessentialprotein-codinggenesinthemetazoanmodelorganismsCaenorhabditiselegansandDrosophilamelanogasteraswellasinMusmusculusandHomosapiens(aHepG2cellline)byintegratingLLMandGNNswithadversarialtraining.Bingopredictsessentialgenesundertwo‘zero-shot’scenarioswithtransferlearning,showingpromisetocompensateforalackofhigh-qualitygenomicandproteomicdatafornon-modelorganisms.Inaddition,theattentionmechanismsandGNNExplainerwereemployedtomanifestthefunctionalsitesandstructuraldomainwithmostcontributiontoessentiality.Inconclusion,Bingoprovidestheprospectofbeingabletoaccuratelyinfertheessentialgenesoflittle-orunder-studiedorganismsofinterest,andprovidesabiologicalexplanationforgeneessentiality.
• 论文类型：期刊论文
• 文献类型：J
• 是否译文：否
• 发表时间：2024-01-12
• 收录刊物：SCI