当前位置:首页 > 报告详情

ChemComp:基于化学反应网络计算的编译框架.pdf

上传人: 芦苇 编号:651757 2025-05-01 22页 2.17MB

1、ChemComp:A Compilation Framework for Computing with Chemical Reaction NetworksAntonino TumeoComputer Scientist,PNNLAuthors:Nicolas Bohm AgostiniConnah G.M.JohnsonWilliam CannonAntonino Tumeo268 kJ/mol in one secondPotential of Biochemical PathwayOPS:1023Energy/s:68 kWFrontier(ORNL,2021)FLOPS:1018Ene

2、rgy/s:29,000 kWRules are simple compared to biologyPotential Gain in OPS/W4x107 xGlucosePyruvateATPADPATPADPATPATPATPADPADPADPGlucose+2 ATP 2 Pyruvate+6 ATP“Chemical advantage”meets the high-performance requirements for scientific workflows:-Scalable-Massively concurrent-Continuous(time)representati

3、on-“Low”energyChemistry conveys a“chemical advantage”for computationBeneficial to applications:-Continuous systems such as ODEs as natural representation-Neural networks/NLP as massively concurrent inferenceChemical reactions can be used for useful computation3Chemical ReactionsOrdinary Differential

4、 EquationsOrdinary Differential EquationsChemical ReactionsMass Action Kinetics LawCompilerD.Soloveichik,et.al.-Turing completeness Natural Computing 2008M.Matejak,et.al.Modelic Lib of Chem Processes Modelica Conf 2015W.Poole,et.al.-BioCRNpyler PLOS Computational Biology 2022R.Rendal,et.al.-Chemilan

5、g Github 2023S.Shah,et.al.-CCompiler arXiv 2008M.Vasic,et.al.-CRN+Natural Computing 2020RW does not consider:Concrete chemical reactionsTracking species over time4Chemical reaction can be mapped to ODEs()=():Number of input species in reaction :Number of output species in reaction():Propensity or Ra

6、te of the reaction Using the Law of Mass Action kinetics:Chemical concentrationTimek1A+BCCDr1:r2:k25Simulated TimeProblem SolutionOne reactionTwo reactionsThree reactionsFour reactionsa)c)b)5Fitting mathematical patterns into concrete CRs6Reservoir computing can harness real world systems for comput

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文介绍了一个名为ChemComp的编译框架,该框架能够将化学反应网络(CRN)映射为普通微分方程(ODEs),从而在生物化学系统中实现高效的计算。ChemComp利用化学反应的“化学优势”,在处理连续系统如ODEs和大规模并行推理如神经网络和自然语言处理(NLP)时表现出优势。文章中提到,通过将化学反应映射为ODEs,ChemComp能够处理高达68 kJ/mol的化学反应能量,并在一秒钟内完成。此外,ChemComp还能在科学工作流中实现可扩展、大规模并行计算,同时保持时间的连续性。文章还讨论了将化学反应网络抽象化为反应和网络的通用形式,并通过编译器支持这些抽象反应的实例化。作者们提出了一种基于微流控技术的生物化学计算设备,并展示了如何使用编译器来研究生物化学系统。最后,文章强调了在考虑CRNCM的可行性和扩展性时,需要考虑物理和化学的约束。
化学计算的新框架" 如何改变计算游戏规则?" 科学计算的革新"
客服
商务合作
小程序
服务号
折叠