CHANGELOG.TXT

5-14-2013 Updates
-----------------
1. Added GCTX write support


11-18-2012 Updates
-------------------
1. Support for GCTX format 

The L1000 project uses a binary format based on the open source HDF5
format, to store matrix data.  This format has been designed as a
drop-in replacement to the commonly used text-based GCT format and
provides efficient random access subsets of large matrices. We've
added Matlab and python routines to parse GCTX files. Additionally
open-source drivers in a number of languages are available at
www.hdfgroup.org/hdf5.

Code:
matlab/parse_gctx.m
python/code_snippets/example_gctx_methods.py

2. Improvements to peak detection

The standard peak detection algorithm (dpeak) assigns values to each
gene pair associated with an analyte based on the readings from single
sample. Improvements in peak detection accuracy can be obtained by
examining the distributions of expression and bead counts for each
pair of genes across all samples in a detection plate. 

This can be regarded as a classification problem with noisy
inputs. Assuming that the expression and counts were determined
correctly for the majority of wells in the plate and that "reciprocal
flips" between the two unrelated genes are unlikely, one can build a
classifier that identifies and corrects 'reciprocal flips'.

Code:
matlab/code_snippets/iterative_flip_adjust_2d.m
matlab/code_snippets/flip_correction_nd.m

3. Signature generation using robust z-scores

A robust zscore metric is used to define pertubational signatures from
L1000 data. Two variants are currently supported:

a) plate population based reference: the zscore is computed relative
to all samples measured in a plate

b) vehicle based reference: the zscore in computed relative to a group
of designated vehicle treatment samples. For example DMSO treated
wells in the case of compound treatments.

Code:
matlab/code_snippets/robust_zscore.m

2012-Feb-29: Updates
---------------------
New tools:
    Normalization routines
	l1kt_liss.m: Apply L1000 invariant set scaling
	l1kt_qnorm.m: Perform quantile normalization
    Inference
	l1kt_infer.m: Apply inference model to a set of normalized landmark profiles.
Renamed 3 tools:
	dpeak -> l1kt_dpeak
	parse_lxb -> l1kt_parse_lxb
	plot_peaks -> l1kt_plot_peaks
Additions to the data folder:
	epsilon_cal.gmx : list of invariant genesets used for normalization.
	log_ybio_epsilon.gct : Affymetrix reference values for invariant genesets
	HG_U133A.chip : Probeset to gene mappings

2011-Dec-31: Initial Release
-----------------------------