DREAM4_Challenge2_GoldStandards/README.rtf

{\rtf1\ansi\ansicpg1252\cocoartf949\cocoasubrtf460
{\fonttbl\f0\fswiss\fcharset0 Helvetica;}
{\colortbl;\red255\green255\blue255;\red0\green0\blue255;}
\paperw11900\paperh16840\margl1440\margr1440\vieww8780\viewh8660\viewkind0
\pard\tx566\tx1133\tx1700\tx2267\tx2834\tx3401\tx3968\tx4535\tx5102\tx5669\tx6236\tx6803\ql\qnatural\pardirnatural

\f0\b\fs28 \cf0 Readme
\b0\fs24 \
\
This archive contains the gold standards of the three subchallenges of the DREAM4 
\i in silico
\i0  network challenge. A description of the challenges can be found on the DREAM website:\
\
\cf2 \ul \ulc2 http://wiki.c2b2.columbia.edu/dream/index.php/D4c2\
\cf0 \ulnone \
Additional information is available on the GeneNetWeaver (GNW) website (\cf2 \ul \ulc2 http://gnw.sourceforge.net\cf0 \ulnone ), including:\
- additional gene-expression datasets\
- protein concentration datasets\
- the datasets without noise\
- the actual names of the genes in e.coli and yeast\
- the perturbations that were applied to generate the time series and the multifactorial perturbation datasets\
- images of the network graphs\
- and more...\
\
\

\b Size 10 and Size 100
\b0 \
\
These two directories contain the gold standards for the 
\i size 10
\i0  and 
\i size 100 
\i0 subchallenges. The files 
\i *goldstandard.tsv 
\i0 contain the true network structures (i.e., a list of edges that are present in the network). \
\
The gold standards for the bonus round can be found in the 
\i Bonus round 
\i0 subdirectory. The files 
\i *nonoise_dualknockouts.tsv
\i0  contain the steady-states of the double knockouts that were asked to be predicted in the bonus round. This data is noise-free: it was simulated using ordinary differential equations (and not stochastic differential equations as for the training data) and no experimental noise (measurement error) was added.\
\
Note that we also evaluated the submitted predictions using a gold standard with the same type of noise as in the training data. This had a negligible effect on the results and did not affect the ranking of teams. Noisy dual-knockout datasets are available in the additional information mentioned above.\
\
Remember that the indexes of the genes that were knocked out in each double-knockout experiment are given in the files 
\i *dualknockouts_indexes.tsv
\i0 , which are part of the training dataset available at: \cf2 \ul \ulc2 http://wiki.c2b2.columbia.edu/dream/data/DREAM4\cf0 \ulnone .\
\
\

\b Size 100 Multifactorial\

\b0 \
The files *goldstandard.tsv contain the true network structures, as mentioned above.\
\
The gene networks of this subchallenge are identical to the ones of the 
\i Size 100 
\i0 subchallenge, we just randomly renamed all the nodes. The files 
\i *goldstandard_ref.tsv 
\i0 in the directory called 
\i Mapping
\i0  contain the mapping of gene names. The original gene names in the 
\i Size 100 
\i0 subchallenge are given in the second column, and the corresponding gene names in the 
\i Size 100 Multifactorial
\i0  subchallenge are given in the first column.\
\
\

\b Questions / Feedback
\b0 \
\
For questions or feedback concerning the 
\i in silico
\i0  challenge, please use the DREAM discussion forum (\cf2 \ul \ulc2 http://wiki.c2b2.columbia.edu/dream/discuss\cf0 \ulnone ) or contact Daniel Marbach (\cf2 \ul \ulc2 dmarbach@mit.edu\cf0 \ulnone ).\
\
\
---\
Daniel Marbach\
21 Nov 2009}