Alexey Uvarovskii. pulseR for RNA kinetics

Analysis of RNA kinetics
using

Alexey Uvarovskii, Christoph Dieterich

University Hospital Heidelberg

Acknowledgement

Tobias Jakobi (Dieterich Lab) - computing support

David Vilchez, Seda Koyuncu (CECAD Cologne),
Janine Altmüller, Marek Franitza (CCG Cologne) - experimental data

Is there a difference?

Number of motor vehicle deaths in the US (wiki)

Same number, different context

I bet you think about

Rates

Reads in RNA-seq

Number of motor vehicle deaths in the US (wiki)

RNA level is a balance

$\sf \frac{d[\text{RNA}]}{dt} = +$ $\sf [\text{synthesis}] $ $\sf -$ $\sf [\text{degradation}]$ $\sf \cdot [\text{RNA}]$

$$ \sf [\text{steady state RNA}] = \frac{[\text{synthesis}]}{[\text{degradation }]}$$

Nascent RNA can be traced

EU = 5-ethyniluridine


from the Click-iT® Kit Thermofisher manual
usually followed by RNA-seq

Pulse-chase experiment

a way to measure RNA kinetics

Background

  • rates are also interesting
  • new RNA can be traced
  • pulse-chase RNA-seq

Analysis

  • kinetic model
  • stat model
  • normalisation

pulseR to help

Alexey Uvarovskii, Christoph Dieterich; pulseR: Versatile computational analysis of RNA turnover from metabolic labeling experiments. Bioinformatics 2017 btx368. doi: 10.1093/bioinformatics/btx368

Kinetic model

defined by the setup, e.g. pulse labelling is

$$ \begin{align} \sf [\text{total}] &= T\equiv\text{const} \\ \sf[\text{pull down}] &= T\cdot \left( 1 - e^{-dt}\right) \end{align}$$

$$\sf t = 1, 2, 4\, \text{hr} $$

Stat model

Negative binomial distribution

purple: no overdispersion, yellow: with overdispersion

Anders, Simon, and Wolfgang Huber. "Differential expression analysis for sequence count data." Genome biology 11.10 (2010): R106.

Normalisation

$$\begin{align}\sf [\text{total}] &= \sf\,[\text{labelled}] + [\text{unlabelled}] \\ \sf[\text{pull down}] &=\sf ?[\text{labelled}] + ?[\text{unlabelled}] \end{align} $$

Normalisation using spike-ins

In pulseR

there are two options for normalisation:

using spike-ins (DESeq)

by MLE fitting

absolute synthesis rate

no spike-ins needed

Alternatives

N: normal, NB: negative binomial, BIN: binomial.

EU pulse-chase on H9 cells

design

A number is not enough

pulseR can estimate confidence intervals for you

diagnostics and comparisons

Confidence intervals

Profile likelihood

$ \frac{\log L(d)}{\log L(d_{optimal})} < \frac{1}{2}\chi^2_{0.95,1 \,d.f.} \approx 1.92 $

Uncertainty in fit

The workflow 

 
library(pulseR)
# put math here
formulas <- MeanFormulas(
  total = mu,
  labelled = mu * (1 - exp(-d*22)) * exp(-d*time),
  unlabelled = mu * (1 - exp(-d*time) * (1 - exp(-d * 22)))
)
# define the fractions
formulaIndexes <- list(
  total_fraction = 'total',
  pull_down      = c('labelled', 'unlabelled'))
 
 
 
pd <- PulseData(counts, conditions, formulas, formulaIndexes,
  groups = ~ fraction + time)
result <- fitModel(pd, initValues, opts)

pulseR allows to

  • estimate kinetic rates from RNA-seq
  • flexible analysis (spike-ins, cross-contamination, etc.)
  • diagnostics with profile likelihood

Poster A-271

An open post-doc position

dieterichlab.org
github.com/dieterich-lab/pulseR

a.uvarovskii@uni-heidelberg.de