Inferring disease-associated lncrnas using expression data and disease-associated protein coding genes

Transcription

1 Inferring disease-associated lncrnas using expression data and disease-associated protein coding genes Xiaoyong Pan Center for non-coding RNA in Technology and Health, Department of Clinical Veterinary and Animal Science, University of Copenhagen Department of Disease Systems Biology, Novo Nordisk FoundationCenter for Protein Research, University of Copenhagen, Denmark. February 20, / 15

2 Introduction 1 lncrnas are emerging as important regulator in different diseases. 2 Gene is more tissue-specific than individual-specific [Melé 2015, Science] 3 Different diseases relevant to specific tissues[lage 2008, PNAS] 4 Disease-associated genes have similar expression pattern. expression value adipose adrenal blood brain Two genes associated with inflammatory bowel disease breast colon heart MADCAM1 in E-MTAB-513 kidney liver lung lymph ovary prostate skeletal_muscle testes thyroid expression value adipose adrenal blood brain breast TAC1 in E-MTAB-513 colon heart kidney liver lung lymph ovary prostate skeletal_muscle testes thyroid Tissue 2 / 15

3 Project goal 1 Infer disease-associated lncrnas from protein coding and lncrna co-expression dataset. 2 Method: Train random forest model on disease-associated protein coding gene expression profiles, then predict for lncrna. 3 / 15

4 Dataset 1 3 RNAseq datasets: - E-MTAB-513, human body map, 16 tissues, GENCODE v7. - GSE43520, evolution of lncrna in tetrapods, 15 tissues, highly conserved lncrnas. - GSE30352, evolution of genes in mammalians, 6 tissues, Ensembl based annotation. [Derrien 2012, Genome Res], [Necsulea 2014, Nature], [Brawand 2011, Nature] 4 / 15

5 Dataset 1 DISEASES [Pletscher-Frankild 2015, Methods] database for disease-associated protein coding genes ,405 associations between 17,606 genes and 4,610 diseases. 2 LncRNADisease [Chen 2013, Nucleic Acids Res] for verified disease-associated lncrnas. - More than 1,000 association between 321 lncrnas and 221 diseases. 5 / 15

6 Infer disease-associated lncrna from co-expression profile 6 / 15

7 Constructing training data 1 For each disease X: - Randomly sampling 5N mrnas subsets not involved in disease X with 5 times the number of mrnas associated with disease X. - assign label 1 to disease-associated mrnas, otherwise 0. 2 NOTICE: when randomly select negative mrnas, this mrnas should be associated with other diseases (well studied genes instead of rare genes), have no any evidence for disease X in DISEASES database. 7 / 15

8 Infer disease-associated lncrna from co-expression profile 8 / 15

9 Performance using mrna expression profile Table: Average performance for diseases, whose # of associated mrnas overlapping with mrnas in expression data is greater than 25 in DISEASES database, using mrna expression profiles. Dataset # diseases Accuracy Sensitivity Specificity Precision MCC E-MTAB GSE GSE / 15

10 Tissue importance for disease-associated gene classification disease-tissue association score[lage 2008, PNAS]. Random forest feature importance analysis. Expression value in tissues are features, tissue important score is ranked by random forest. 90 Prostate cancer 0.09 Prostate cancer Association score Importance score (Adrenal Cortex) (B cells) (B lymphoblasts) (CD4 cells) (CD8 T cells) (Cardiac Myocytes) (Cerebellum Peduncles) (Colorectal Adenocarcinoma) (Medulla Oblongata) (Monocytes) (NK Cells) (Occipital Lobe) (Olfactory Bulb) (Pancreatic Islets) (Parietal Lobe) (Prefrontal Cortex) (Skeletal Muscle) (Smooth Muscle) (Temporal Lobe) (Testis Germ Cell) (Testis Leydig Cell) (Testis Seminiferous Tubule) (Trigeminal Ganglion) (WHOLE BLOOD) (Whole Brain) (adrenal gland) (atrioventricular node) (bone marrow CD34) (bone marrow) (caudate nucleus) (ciliary ganglion) (dorsal root ganglion) (fetal Thyroid) (fetal brain) (fetal liver) (fetal lung) (globus pallidus) (leukemia chronic myelogenous) (leukemia lymphoblastic) (leukemia promyelocytic) (lymph node) (salivary gland) (spinal cord) (subthalamic nucleus) Adipocyte Amygdala Appendix Heart Hypothalamus Liver Lung Ovary Placenta Pancreas Pituitary Pons Prostate Tongue Thalamus Thyroid Tonsil Uterus cerebellum kidney skin testis thymus trachea adipose adrenal blood brain breast colon heart kidney liver lung lymph ovary prostate skeletal_muscle testes thyroid Lage 2008, PNAS Tissue 10 / 15

11 Infer disease-associated lncrna 1 For each association in LncRNADisease, randomly select another lncrna for this disease as negative pair. 11 / 15

12 Benchmarking predicted disease-association lncrnas A B disease-lncrna pairs lncrna in gold standard FP TP disease in gold standard 1 precision wrt. gold standard 0 sliding window predicted raw score in window w raw interaction score 12 / 15

13 Benchmark against disease-lncrna from LncRNADisease 0.20 E-MTAB-513_lncRNA.tsv 0.16 GSE30352_lncRNA.tsv Probability Probability Raw score Raw score 0.14 GSE43520_lncRNA.tsv Probability Raw score 13 / 15

14 Conclusion and outlook 1 In this study, we infer disease-associated lncrna from expression data and disease-associated protein coding genes. 2 Integrate GWAS SNP data with predicted score to prioritize disease-associated lncrnas. 3 Text mining disease-lncrna associations and compared our prediction to it. 4 How to better select negative genes for model training. 14 / 15

15 Acknowledge! Lars Juhl Jensen Jan Gorodkin RTH and DSB group Funding: Innovation fund Denmark PhD scholarship from from Faculty of Health and Medical Science, University of Copenhagen Thanks for your attention! 15 / 15