Journal: BMC Genomics

Article Title: Integrative modeling of lncRNA-chromatin interaction maps reveals diverse mechanisms of nuclear retention

doi: 10.1186/s12864-023-09498-9

Figure Lengend Snippet: Model performance and feature family analysis. a Bar chart illustrates the test performance of the RF classifier using all features, for each lncRNA in random/block cross-validation settings. b and c illustrate PR and ROC curves respectively, quantifying the test performance of RF models trained to predict Gm14820-chromatin interactions. d Schematic representation of the Marginal Predictive Value (MPV) evaluation framwork. The toy example is a scenario with two primary and two secondary families of features. The MPV of a feature family is quantified as the average difference in test performance of pairs of models that have been trained using feature sets that only differ in the inclusion of that feature family (see ). e Boxplots (median and quartiles) show the contributions of feature families to the prediction of Malat1-chromatin interactions. Each boxplot corresponds to one feature family and depicts the distribution of 16 points representing pairwise differences in performance of models differing only in inclusion of that particular feature family. f Heatmap shows MPV of feature familes (columns) for all lncRNAs (rows). Row label colors show lncRNA group (TA and non-TA lncRNAs shown in gold and navy, respectively). g Black line shows the test ROC of a classifier that uses sequence features only, whereas the green line shows the test ROC of a classifier that uses the transcription family of features in addition to sequence features to predict Trerf1-chromatin interactions. h ROC curve shows the performance of a classifier that uses triplex feature family to predict chromatin interactions for Rian. i Boxplots illustrate the MPV for Triplex and Transcription feature families grouped by lncRNA type. * indicates Mann–Whitney p -value of 0.01. j ROC curves showing performance of models trained to predict Gm14820 chromatin interactions. k Average complementarity scores for 16 top-scoring pairs of feature families are shown. Chrom: Chromatin marks; Methyl: Methylation. l ROC curves of models trained to predict Neat1-chromatin interactions using DNA methylation alone (black), Histone marks alone (green), or both (blue) as predictors. m Boxplots indicate the performance of models trained using methylation and chromatin mark features for all lncRNAs grouped by their type

Article Snippet: Chrom: Chromatin marks; Methyl: Methylation. l ROC curves of models trained to predict Neat1-chromatin interactions using DNA methylation alone (black), Histone marks alone (green), or both (blue) as predictors. m Boxplots indicate the performance of models trained using methylation and chromatin mark features for all lncRNAs grouped by their type

Techniques: Blocking Assay, Biomarker Discovery, Sequencing, MANN-WHITNEY, Methylation, DNA Methylation Assay