ادغام داده براي پيش بيني يال هاي رابط بين سطوح اميكس در شبكه هاي چند لايه ميانكنشي بر اساس تجزيه ماتريسي

چكيده انگليسي :

Data integration for prediction ofinter omics layers interactions in multi layer networks using matrix factorization Niloofar Borhani n borhani@ec iut ac ir January 19 2019 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisors Dr Jafar Ghaisari Dr Marzieh Kamali Advisor Yousof GheisariAbstract New technologies and methodologies in biological systems make the measurement of more characteristics of cellularcomponents such as genes proteins transcripts and metabolites possible These components have essential interactionsamong each other A comprehensive evaluation of functions and communications of each cellular component in systemsbiology is called omics the most important of which are layers of genomics proteomics transcriptomics and metabolomics These omics layers are related to each other by fundamental interactions So far due to the extent and complexity ofbiological systems the investigations have been focused on one of the layers while the interactions between those layers havenot been taken into account Results of studying one omics layer describe a limited part of a biological system however many diseases originate as a result of interactions in broad and complex molecular network Therefore the interactionsof omics and effects of omics layers on each other should also be considered in biological systems investigations Theintegration of omics data would lead to a better understanding of functions of cellular components cause of disease andidentification of drug targets In this thesis omics data has been integrated and a computational method has been developedfor predicting interactions between omics layers in multi layer interaction networks With the development of variousforms of matrix factorization methods and the emergence of deep learning the integration of information and data of theomics layers as well as the creation a multi layer network of omics interactions has been made possible In this thesis miRNAs and proteomics data in Diabetic nephropathy collected from real experiments has been used In order to predict theinteractions between proteins and miRNAs a generalized non negative matrix tri factorization method has been proposed The proposed model has employed methods on the interaction matrix of these layers This model is not only reduces largesize and complexity of data but also reveals latent components of data Then protein interactions and gene ontology datahas been integrated to analyze effects of data integration Because of the dynamic nature of biological phenomena andthe probabilistic characteristics of experimental measurements integration of data limits model performance Then deepmatrix factorization method has been used which improved model performance In this method two deep neural networksfor proteomics and miRNAs are used to find the best representation vectors for proteins and miRNAs Also employinga decoder has improved model performance By modifying inputs or adding neural networks a generalization of deepmatrix factorization method has been proposed to integrate more information in the model as well as making the modelingof more than two heterogeneous object types possible The problem of over fitting in deep neural networks has been solvedby techniques such as dropout regularization and early stopping It also has been suggested to use the singular valuedecomposition technique to determine the dimensions of the representation vectors To evaluate the performance of themethods two sets of colon cancer data and Gene ontology of the genes have been modeled In this thesis a novel method ofmodeling multi layer interaction network has been proposed to investigate omics layers together which will help to predictmore appropriate drug targets in the future Key Words Link prediction Matrix factorization Deep matrix factorization Multi layer interactionnetwork Omics integration Data integration

برهاني، نيلوفر

ادغام داده براي پيش بيني يال هاي رابط بين سطوح اميكس در شبكه هاي چند لايه ميانكنشي بر اساس تجزيه ماتريسي

پيش‌بيني يال , تجزيه ماتريسي , تجزيه ماتريسي عميق , شبكه چند لايه ميانكنشي , ادغام اميكس‌ها , ادغام داده