Interpretation of ‘Omics dynamics in a single subject using local estimates of dispersion between two transcriptomes

Abstract

Calculating Differentially Expressed Genes (DEGs) from RNA-sequencing requires replicates to estimate gene-wise variability, infeasible in clinics. By imposing restrictive transcriptome-wide assumptions limiting inferential opportunities of conventional methods (edgeR, NOISeq-sim, DESeq, DEGseq), comparing two conditions without replicates (TCWR) has been proposed, but not evaluated. Under TCWR conditions (e.g., unaffected tissue vs. tumor), differences of transformed expression of the proposed individualized DEG (iDEG) method follow a distribution calculated across a local partition of related transcripts at baseline expression; thereafter the probability of each DEG is estimated by empirical Bayes with local false discovery rate control using a two-group mixture model. In extensive simulation studies of TCWR methods, iDEG and NOISeq are more accurate at 5%<DEGs<20% (precision>90%, recall>75%, false_positive_rate<1%) and 30%<DEGs<40% (precision=recall~90%), respectively. The proposed iDEG method borrows localized distribution information from the same individual, a strategy that improves accuracy to compare transcriptomes in absence of replicates at low DEGs conditions.

binomialRF: interpretable combinatoric efficiency of random forests to identify biomarker interactions

“…Strobl and Zeileis [30] demonstrate that i) the Gini importance (measure of entropy) is biased towards predictors with many categories, and ii) that growing more trees inflates anticonservative power estimates. To address (i), we recommend the user evaluates sets of genes according to their baseline expression levels [31]. For the latter (ii), the binomialRF uses ntree parameter (number of trees; Table 6) to calculate a conservative cumulative distribution function (cdf) rather than calculating an anticonservative F j (Eq.…”

“…Some algorithms identify sets of conditional or sequential splits, while other strategies (i.e., [37]) measure their effect in prediction error. More recently, works such as [31,38] look at the frequency of sequence of splits or "decision paths" as a way to determine whether two features interact in the treesplitting process. For example, iterative random forests (iRF) [38] identify decision paths along random forests and captures their prevalence, therefore benefitting from a combinatoric feature space reduction in the interaction search.…”

Evaluating single-subject study methods for personal transcriptomic interpretations to advance precision medicine

“…As simulations and synthetic data can investigate a ranges of accuracies against a true gold standard, though they can be prone to other biases and limitations. Li et al (14) have implemented a comprehensive simulation of ss-DEG methods across 8000 tests in a companion study, using a range of DEG proportions from 5-40%, assuming distinct distributions (Poisson or negative binomial), and modeling a variable mean to variance relationship observed from real datasets as recommended by McCarthy et al (32). The results from those simulations broadly agree with the results obtained in this study, identifying the same precision and recall rankings between NOISeq-sim, edgeR, DESeq, and DEGSeq.…”

“…However, transcriptional dynamics operating and validated at the gene set or pathway-level cannot straightforwardly be deconvoluted to identify specific transcripts altered in a single-subject. A recent study provides a comparison of accuracy of five ss-DEGs methods using computer simulations of several data models with genomic dysregulation ranging from 5 to 40% DEGs (14). A partial biological validation was conducted for one ss-DEG methods, NOIseq (15), confirming the top 400 DEG signals by qPCRs.…”

binomialRF: Interpretable combinatoric efficiency of random forests to identify biomarker interactions

“…Strobl and Zeileis [38] demonstrate that i) the Gini importance (meas-ure of entropy) is biased towards predictors with many categories, and ii that growing more trees inflates anticonservative power estimates. To address (i), we recommend the user evaluates sets of genes according t their baseline expression levels [39]. For the latter (ii), the binomialRF uses ntree parameter (number of trees; Table 2) to calculate a conserva tive cumulative distribution function (cdf) rather than calculating a anticonservative (Eq.…”