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

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Survival regression by data fusion

Talk at CAMDA, ISMB'14, Boston, MA, USA.
by

Marinka Zitnik

on 14 August 2014

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Survival regression by data fusion

Survival regression by data fusion
Survival regression with collective matrix factorization
Data
Effects of data features on patient survival time
Motivation
Marinka Zitnik & Blaz Zupan
University of Ljubljana, Slovenia
Baylor College of Medicine, Houston, USA
1
1,2
1
2
Hofree
et al
. 2013
Aims
Can
be done jointly?
and
Aalen's additive model
Joint prediction and estimation through optimization
Experimental setup
3 ICGC cancer cohorts
Head and neck squamous cell carcinoma
Lung adenocarcinoma
Kidney renal clear cell carcinoma
Data fusion graph for one cancer
Survival time prediction
Joint inference of fused latent and survival models
Data fusion and joint inference contributed to improved performance
Information about changing effects of latent factors on survival time
Genes associated with latent factors related to known cancer processes
Regularization
Survival regression
Simultaneous data fusion and survival regression
Application to ICGC cancer cohorts
Model survival time as a function of gene, protein and miRNA expression, methylated regions and mutations
Survival analysis
Survival regression
Survival function
Probability that the event has not occurred yet at time
Hazard function
Probability of event happening the next instant given it hasn't happened yet
Estimate survival function non-parametrically from the data
Patient is right-censored if the event has not yet occurred by the end of the study
Estimating survival function using Kaplan-Meier
the number of events at time
the number of patients at risk of event at time
leave-one-out cross-validation of tumor samples
Log-transformed absolute error loss of survival time
is predicted median of survival time
Optimal predictor of the absolute error loss
Less affected by the long tails of survival distribution than the squared error loss
Error at a distant time point contributes less
than the same error at a nearer time point
Lawless an Yuan 2010
Estimating hazard rates using Nelson-Aalen
cumulative hazard function estimator
Patient
lifetime data
Specific data at the patient level
Rate of change of the curve is estimate of hazard function
Data fitting
hazard rates of patients at
feature profiles of patients at risk at
time-varying
regression
coefficients
Object type
Data set
Backbone
matrix
Recipe matrix
Survival data
Toy example of survival
regression by data fusion
time-varying survival
regression coefficients
Latent factors obtained by collective matrix tri-factorization
Recipe matrix
Backbone matrix
Time-varying survival
regression coefficients
Survival time prediction
Regress protein expression data against survival data
Sequential
survival regression
Original data domain
Regress somatic mutations
against survival data
Inference of latent space = data
dimensionality reduction
Months
Cumulative hazard
Months
Months
Regression coefficient
HNSC cancer study
starts off small in
the first 10 months
after primary diagnosis
HNSC cancer study
Regression coefficient
Latent factor 1
Latent factor 2
different dynamics for
the two latent factors
Genes associated with latent factors are
enriched in processes with roles in HNSC
Genes from latent factor 2
Genetic interaction
Co-expression
Bourguignon
et al.
2006
Bild
et al
. 2006
Burington
et al
. 2008
Lim
et al
. 2012
Mallon
et al
. 2013
Negative regulation of
cyclase activity
Regulation of cAMP
biosynthetic process
Striated muscle hypertrophy
Actin cytoskeleton
organization
Regulation of cardiac
muscle hypertrophy
Data fusion graph
Co-localization
FDR < 0.0001
Assignment of genes to latent factor
Survival regression coefficients of latent factor 2
Latent factor 2
Gene membership of
latent factor 2
Tumor samples associated with survival data
"donor's vital status"
"donor's interval of last followup"
Gene Ontology
annotations
Protein expression
miRNA expression
Gene expression
Mutated genes
in the tumor
Gene-based Beta
values of
interrogated sites
Conclusions
Methods highlight
Aalen 1989
Aalen 1993
Abadi
et al
. 2011
Baseline coefficient
Survival regression with collective matrix factorization
Additive model with time-varying regression coefficients
Survival analysis
ICGC cohorts
Survival time prediction
Collective model
Possible utility for uncovering critical factors and their changing influences across stages of cancer progression
Acknowledgments
Travel fellowship
Principle of collective matrix factorization
(
)
and
without
#F40
Gene prioritization
Gene function prediction
Disease-disease associations
Drug interaction prediction
Drug toxicity prediction
Applied so far:
Iterative multiplicative update rules
Update
,
with rules from Zitnik & Zupan
,
and
Generalized linear matrix equation
Reduction to
Sylvester equation
Update
and
:
Full transcript