Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Lab Meeting:

No description
by

Elisa Margolis

on 19 January 2016

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Lab Meeting:

7 women
3-20 time series
fail to distinguish between interactions & similar response to environment
Lab Meeting:
Microbiome Causality Conundrum

October 19 2015
Ellie Margolis
Why is causality so difficult
in microbiome studies?
The Community You END uP with
community
constantly
shifting
individuals react to environment, disturbances & other species
examples:
a) response to antibiotics: disease community abnormal due to having disease or to antibiotics/treatments for disease
b) graft v. host: immune response triggered by abnormal community or abnormal immune response responsible for community
c) hepatic disease- peritonitis: abnormal immune response triggers dysbiosis from which peritonitis org or hepatic biliary abnormality responsible for dysbiosis then responsible for abnormal immune response
d) bacterial vaginosis: dysbiosis triggered by loss of lactos (leads to rise of bvab) or rise of bv species (leads to loss of lactos)
e) malnutrition (or obesity): changes in gut bacteria reduce (or increase) efficiency of community metabolism or does changes in diet select for bacteria with reduced (or increased) metabolic efficiency

Going beyond patterns...
... to investigate processes
absence/presence
community assembly rules
dynamic
Correlations
Granger Causality
PREMiSE: cause must precede event
evidence for one species' improving forecast (predictive causality) of another species' time series
Some discernment
Interactions v Forcing
Positive v Negative
Results: Vaginal Microbiome
Results: Vaginal Microbiome
Steps in Granger Causality
Why Granger Causality isn't sufficient?
Caveats & Limitations
(pairwise-other confounders, short duration, mis-specified lag lengths, insufficient sampling relative to noise)
Invasions/Disturbances
Cross Convergent Mapping (CCM)
PREMISE: Cause Precedes Event
measure extent historical record's species Y estimates states of species X
signature of X in Y's time series
adapted from Sugihara et al. Science (2012) 338:496-500
(state space embedding; dynamic attractor- linear & nonlinear)-
embedding dimension
predictive value
causality testing
Cross Convergent Mapping
Advantage: semi-quantitiative data
Major Advantage:
Combine multiple short time series
now we can ask....
what interactions occur in bv & non-bv communities?
can changes in specific species be linked to changes in bv scoring?
do interactions change during recovery from antibiotics?
which species modify those interactions?
flora v. bv-like
no antibiotics or recovery (7 days after antibiotics), exclude more than 3 continuous days at lower limit of detection
L. iners Dynamics
Preliminary:
Interaction Modifiers
natural reductionist experiments
next steps
interaction modifiers: co-culture experiments
analytical tool development: disturbances, conditional rare species
potential L. iners investigations into distinct dynamics
validate further semi-quantitative data
(next gen sequencing)
hard to integrate different levels
(species, strains, metabolites, genes)
synchrony
spurious
correlations
environmental
forcing
species
interactions
shared noise (sampling)
Correlations of Dynamics
Daily Relative Frequency change
Correlations of Dynamics
log daily differences qPCR
1) Determine Lag Length
number of historic
values used to forecast
Likelihood Ratio to optimize number of prior values being accounted
Constructing vector autoregression model
2) Error Correction term (cointegration)
shared sampling noise
3) Test whether inclusion of second species improves forecastabiliy of model
unidirectional
bidirectional
microbe-microbe
interaction
environmental
forcing
confounder
microbe-microbe
interaction
confounder
Granger Causality:
qPCR (short 20 day segment)
Granger Causality:
pyro (short 20 day segment)
Uncertainty
sampling
(qPCR)
sampling
(swab &
extraction)
sampling
(swab &
extraction)
pipelines
errors
(classification, sequencing)
and other data types (gene counts, metabolites, patient scoring, etc)
CCM: qPCR (full 58 days)
CCM: pyro (56 days)
6 women, time series 3-16 days
17 women, time series 3-30 days
Library Length
(as you fill in the
manifold)
(more likely to find
one species influence
on another)
Pearsons correlation
10 women
4-10 days time series
interplay between
host & community
antibiotics
OR the illness they were treating
explore interpretations:
Correlation
Granger Causality
Cross Convergent Mappimg
Data attributes
Semi-quantitative (ngs)
Coupling
(classification)
More you know about the cause the better you can predict the response
Correspondence between library of points of Y's attractor manifold and X's
Prediction Interval
flora
bv-like
how good is the mapping at forecasting into the future?
(diagnoses of dynamical patterns)
9 women
3-30 time series
Context Questions
Full transcript