LINA: A Linearizing Neural Network Architecture for Accurate First-Order and Second-Order Interpretations.
LINA: A Linearizing Neural Network Architecture for Accurate First-Order and Second-Order Interpretations.
复制标题
DOI：
10.1109/access.2022.3163257
复制
发表时间：
2022
期刊：
影响因子：
3.9
通讯作者：
Pan, Chongle
中科院分区：
文献类型：
--
作者：
Badre, Adrien;Pan, Chongle
关键词：
虽然神经网络可以提供高预测性能，但识别用于其预测的显著特征和重要特征交互是一个挑战。这是在许多需要可解释性的生物医学应用中部署神经网络的关键障碍，包括预测基因组学。在本文中，线性化神经网络架构（LINA）在这里开发提供的一阶和二阶解释的实例和模型的水平。LINA结合了深层内在注意力神经网络的表示能力和用于模型解释的线性化中间表示。与DeepLIFT、LIME、格拉德 *Input和L2X相比，LINA的一阶解释与合成数据集中特征的地面真实重要性排名具有更好的斯皮尔曼相关性。与NID和GEH相比，LINA的二阶解释结果在识别合成数据集的地面实况特征相互作用方面具有更好的精度。这些算法进一步基准使用预测基因组学作为现实世界的应用。在给定的错误发现率下，LINA比其他算法识别出更多的重要单核苷酸多态性（SNP）和显著的SNP相互作用。结果表明，使用LINA的准确和通用的模型解释。
While neural networks can provide high predictive performance, it was a challenge to identify the salient features and important feature interactions used for their predictions. This represented a key hurdle for deploying neural networks in many biomedical applications that require interpretability, including predictive genomics. In this paper, linearizing neural network architecture (LINA) was developed here to provide both the first-order and the second-order interpretations on both the instance-wise and the model-wise levels. LINA combines the representational capacity of a deep inner attention neural network with a linearized intermediate representation for model interpretation. In comparison with DeepLIFT, LIME, Grad*Input and L2X, the first-order interpretation of LINA had better Spearman correlation with the ground-truth importance rankings of features in synthetic datasets. In comparison with NID and GEH, the second-order interpretation results from LINA achieved better precision for identification of the ground-truth feature interactions in synthetic datasets. These algorithms were further benchmarked using predictive genomics as a real-world application. LINA identified larger numbers of important single nucleotide polymorphisms (SNPs) and salient SNP interactions than the other algorithms at given false discovery rates. The results showed accurate and versatile model interpretation using LINA.
登录
查看更多内容
影响因子：
5.4
作者：
Velasco-Ruiz A;Nuñez-Torres R;Pita G;Wildiers H;Lambrechts D;Hatse S;Delombaerde D;Van Brussel T;Alonso MR;Alvarez N;Herraez B;Vulsteke C;Zamora P;Lopez-Fernandez T;Gonzalez-Neira A
通讯作者：
Gonzalez-Neira A
影响因子：
8.8
作者：
Romualdo Cardoso S;Gillespie A;Haider S;Fletcher O
通讯作者：
Fletcher O
影响因子：
4.2
作者：
Ho Thanh Lam L;Le NH;Van Tuan L;Tran Ban H;Nguyen Khanh Hung T;Nguyen NTK;Huu Dang L;Le NQK
通讯作者：
Le NQK
DOI：
10.1158/1055-9965.epi-16-0106
发表时间：
2017-01
期刊：
Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology
影响因子：
--
作者：
Amos CI;Dennis J;Wang Z;Byun J;Schumacher FR;Gayther SA;Casey G;Hunter DJ;Sellers TA;Gruber SB;Dunning AM;Michailidou K;Fachal L;Doheny K;Spurdle AB;Li Y;Xiao X;Romm J;Pugh E;Coetzee GA;Hazelett DJ;Bojesen SE;Caga-Anan C;Haiman CA;Kamal A;Luccarini C;Tessier D;Vincent D;Bacot F;Van Den Berg DJ;Nelson S;Demetriades S;Goldgar DE;Couch FJ;Forman JL;Giles GG;Conti DV;Bickeböller H;Risch A;Waldenberger M;Brüske-Hohlfeld I;Hicks BD;Ling H;McGuffog L;Lee A;Kuchenbaecker K;Soucy P;Manz J;Cunningham JM;Butterbach K;Kote-Jarai Z;Kraft P;FitzGerald L;Lindström S;Adams M;McKay JD;Phelan CM;Benlloch S;Kelemen LE;Brennan P;Riggan M;O'Mara TA;Shen H;Shi Y;Thompson DJ;Goodman MT;Nielsen SF;Berchuck A;Laboissiere S;Schmit SL;Shelford T;Edlund CK;Taylor JA;Field JK;Park SK;Offit K;Thomassen M;Schmutzler R;Ottini L;Hung RJ;Marchini J;Amin Al Olama A;Peters U;Eeles RA;Seldin MF;Gillanders E;Seminara D;Antoniou AC;Pharoah PD;Chenevix-Trench G;Chanock SJ;Simard J;Easton DF
通讯作者：
Easton DF
DOI：
10.1038/nrg2452
发表时间：
2008-11
期刊：
Nature reviews. Genetics
影响因子：
--
作者：
通讯作者：
--