330 likes | 421 Views
Human Liver Analysis. Alignment + Quantitation. Ensembl 66, Noncode 3.0 Noncodes filters: no Ens overlaps, no Guttman . After filtering out genes with all FPKM < 1: TOTAL: 19,793 (29.4% total) Ensembl PC: 13,612 (66.4% total)
E N D
Alignment + Quantitation • Ensembl 66, Noncode 3.0 • Noncodes filters: no Ens overlaps, no Guttman. • After filtering out genes with all FPKM < 1: • TOTAL: 19,793 (29.4% total) • Ensembl PC: 13,612 (66.4% total) • Ensembl NC: 4,863 (14.7% total) • NoncodeIntronic: 434 (14.6% total) • NoncodeIntergenic: 884 ( 8.2% total) • FINAL NONCODING: 4,870 (78.8% expressed) • No Ensembl t / r / sn / sno / mi / misc-RNAs
ANOVA Significants from 19,793 genes, adjusted p-values ≤ 0.05: Biotypes for significant NC genes:
PC-NC Neighbor Assessment • Nearest PC neighbors for every NC gene found. • Pairs with expressed NC, ≤ 10kb: 2833 • 1747 Intronic, 1086 Intergenic
Clustering • Cluster ANOVA-significant genes only: • Drug: zscores; one avg value per gene per drug • Source: Log2FC: one FC per gene per drug • Interaction: zscores: all values per gene • Z-scores clustered by Pearson correlation dist • Clusters cut from dendrogram • Log2FC clustered by Euclidean dist • For viewing only; actual clusters via k-means
PC Drug NC Drug DMSO PHENO RIFAM DMSO PHENO RIFAM
PC GO / KEGG Terms, raw p ≤ 0.05, 2+ genes/term Drug Effect: Six Clusters
PC Source NC Source DMSO PHENO RIFAM DMSO PHENO RIFAM 724 4388 2036 3724
Source Clusters • GO/KEGG Terms: Adj.p ≤ 0.05, 2+ genes/term: • PC: 1879 up, 1414 down • NC: nothing… Human GO based on UniProt • Neighbor Pairs: Significant trending observed: χ2 p-value ≈ 0
Neighbors with Distance ≤ 10k • Neighbor Pairs: Significant trending observed: χ2 p-value ≈ 0
PC Drug-Source NC Drug-Source PHD PHP PHR RGD RGP RGR PHD PHP PHR RGD RGP RGR
Dif / Pro Log2FPKM, same order Z-Scores
Alignment + Quantitation • Ensembl 66, Noncode 3.0 • Noncodes filters: no Ens overlaps, no Guttman. • After filtering out genes with all FPKM < 1: • TOTAL: 15,244 (27.1% total) • Ensembl PC: 12,990 (56.0% total) • Ensembl NC: 1,516 (10.2% total) • NoncodeIntronic: 290 ( 2.8% total) • NoncodeIntergenic: 448 ( 5.6% total) • FINAL NONCODING: 1,932 (85.7% expressed) • No Ensembl t / r / sn / sno / mi / misc-RNAs
ANOVA Significants from 15,244 genes, adjusted p-values ≤ 0.05: Biotypes for significant NC genes:
PC-NC Neighbor Assessment • Nearest PC neighbors for every NC gene found. • Pairs with expressed NC, ≤ 10kb: 1356 • 766 Intronic, 590 Intergenic
Clustering • Z-scores clustered by Pearson correlation dist • Clusters cut from dendrogram • To distinguish Neonatal, Adolescent, Mature stages, 9 PC clusters and 11 NC clusters required. • Also detected a “Neonatal-Mature” cluster. • NC has 2 types of adolescent clusters; merged.
PC NC -2 0 1 3 5 10 15 20 25 30 45 60 -2 0 1 3 5 10 15 20 25 30 45 60
Time Effect: 4 Main Clusters PC GO / KEGG Terms, adj p ≤ 0.05, 2+ genes/term
Neighbor Pairs • Again, significant trending observed: χ2 p-value ≈ 0
Neighbors with Distance ≤ 10k • Again, significant trending observed: χ2 p-value ≈ 0
Datasets Available • All Genes+FPKMs • Genes+FPKMs per cluster • GO/KEGG Terms per cluster • Concordant neighbor pairs per cluster • For mouse, human source only
GO Themes for Clusters • Human Drug: • 1 Generic Drug Response+Metabolism / * Metabolism • 2 Transcription / Cardiovascular Development? • 3 Cell Adhesion / Steroid Hormone Metabolism • 4 Cell Cycle • 5 Cytoskeleton / Organelle Organization • 6 "Cell Part Morphogenesis" / RNA processing? • Human Drug-Source: • 1 Generic Drug Response+Metabolism, * Metabolism • 2 Organ / Anatomical Structure Development, Cellular Stress Response, Protein Complex Assembly, DNA Metabolism • Human Dif-Pro: • 1 Signaling, Transport, Metabolism, Wnt, MapK • 2 Cell Adhesion/Migration/Localization • Human Source: • Too many • Mouse Time: • Too many
More Human Correlations PC + NC NC PC Scale has been fixed across heatmaps. Within NCs, intronics correlate worse than intergenics.
More Mouse Correlations PC + NC NC PC Scale has been fixed across heatmaps. PC+NC is only slightly better than PC-only.