2,655
edits
Cc analysis: Difference between revisions
m
→Example
m (→Example) |
|||
| (10 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
The purpose of cc_analysis is to find a representation in low-dimensional space (e.g. 2D, 3D, 4D) that displays the relations of data sets to each other. For this, the procedure needs the values of correlation coefficients (CCs) calculated between the data sets (e.g. crystallographic intensities, or pixels of images). (Not all theoretically possible pairwise CCs are needed, but of course, the more the better.) | The purpose of cc_analysis is to find a representation in low-dimensional space (e.g. 2D, 3D, 4D) that displays the relations of data sets to each other. For this, the procedure needs the values of correlation coefficients (CCs) calculated between the data sets (e.g. crystallographic intensities, or pixels of images). (Not all theoretically possible pairwise CCs are needed, but of course, the more the better.) | ||
Since the data sets are composed of many measurements, they could be thought of as residing in a high-dimensional space: in case of crystallography, that dimension is the number of unique reflections; in case of images, the number of pixels. | Since the data sets are composed of many measurements, they could be thought of as residing in a high-dimensional space: in case of crystallography, that dimension is the number of unique reflections; in case of images, the number of pixels. | ||
How is the low-dimensional representation found? If N is the number of data sets and <math>cc_{ij}</math> denotes the correlation coefficients between data sets i and j, cc_analysis minimizes | |||
<math>\Phi(\mathbf{x} )=\sum_{i=1}^{N-1}\sum_{j=i+1}^{N}\left(cc_{ij}-x_{i}\cdot x_{j}\right)^{2}</math> | |||
as a function of the vector <math>\bf{x}</math>, the column vector of the N low-dimensional vectors <math>\it{\{{x_{k}\}}}</math>. This can be performed by minimizing from random starting positions, or more elegantly and efficiently by obtaining starting positions through Eigen decomposition after estimating the missing values of the matrix of correlation coefficients. | |||
As the resulting vectors <math>\it{\{{x_{k}\}}}</math> are in low-dimensional space, and the data sets reside in high-dimensional space, the procedure may be considered as ''multidimensional scaling''. This procedure has first been described in [http://journals.iucr.org/d/issues/2017/04/00/rr5141/index.html Diederichs, Acta D (2017)]. We can also think of the procedure as ''unsupervised learning'', because it "learns" from the given CCs, and predicts the unknown CCs - or rather, the relations of even those data sets that have nothing (crystallography: no reflections; imaging: no pixels) in common. | |||
== Properties of cc_analysis == | == Properties of cc_analysis == | ||
| Line 9: | Line 15: | ||
It turns out that the dot product (also called scalar product, or inner product) of low-dimensional vectors (representing the high-dimensional data sets) is very appropriate for approximating the CCs. This is because a CC is itself a dot product (eqn. 3 of [http://journals.iucr.org/d/issues/2017/04/00/rr5141/index.html Diederichs, Acta D (2017)). | It turns out that the dot product (also called scalar product, or inner product) of low-dimensional vectors (representing the high-dimensional data sets) is very appropriate for approximating the CCs. This is because a CC is itself a dot product (eqn. 3 of [http://journals.iucr.org/d/issues/2017/04/00/rr5141/index.html Diederichs, Acta D (2017)). | ||
It further turns out that | It further turns out that the resulting arrangement of vectors (that minimize the function <math>\Phi(\mathbf{x})</math> given above) has the properties | ||
# 0 <= length <= 1 for each vector; short vectors have a low signal-to-noise ratio, long vectors a good signal-to-noise ratio, and vectors of length 1 represent ideal (prototypical) data sets. In fact, the length of each vector is CC* (as defined in [http://dx.doi.org/10.1126/science.1218231 Karplus & Diederichs (2012)]), and its | # 0 <= length <= 1 for each vector; short vectors have a low signal-to-noise ratio, long vectors a good signal-to-noise ratio, and vectors of length 1 represent ideal (prototypical) data sets. In fact, the length of each vector is CC* (as defined in [http://dx.doi.org/10.1126/science.1218231 Karplus & Diederichs (2012)]), and its relation to the signal-to-noise ratio is given in eqn. 4. | ||
# vectors point in the same direction (i.e. they lie on a radial line) if their data sets only differ by random noise | # vectors point in the same direction (i.e. they lie on a radial line) if their data sets only differ by random noise | ||
# vectors representing systematically different (heterogeneous) data sets enclose an angle whose cosine is the CC of their respective ideal data sets (eqn. 5). | # vectors representing systematically different (heterogeneous) data sets enclose an angle whose cosine is the CC of their respective ideal data sets (eqn. 5). | ||
# if all CCs are known, the solution is unique in terms of lengths of vectors, and angles between them. However, a rotated (around the origin) or inverted (through the origin) arrangement of the vectors leaves the functional unchanged, because these transformations do not change lengths and angles. | # if all CCs are known, the solution is unique in terms of lengths of vectors, and angles between them. However, a rotated (around the origin) or inverted (through the origin) arrangement of the vectors leaves the functional unchanged, because these transformations do not change lengths and angles. | ||
# as long as the problem is | # as long as the problem is well-determined, the vectors can be calculated. Unknown CCs between data sets (e.g. in case of crystallographic data sets that don't have common reflections) can be estimated from the dot product of their vectors. Well-determination means: each data set has to be related (directly or indirectly i.e through others) to any other by at least as many CCs as the desired number of dimensions is. A necessary condition for this is that each data set has at least as many relations to others (input lines to cc_analysis involving this data set) as the number of dimensions is. It is of course better if more relations are specified! | ||
== The program == | == The Fortran program == | ||
<code>cc_analysis</code> calculates the vectors from the pairwise correlation coefficients. The (low) dimension must be specified, and a file with lines specifying the correlation coefficients must be provided. | <code>cc_analysis</code> calculates the vectors from the pairwise correlation coefficients. The (low) dimension must be specified, and a file with lines specifying the correlation coefficients must be provided. | ||
| Line 23: | Line 29: | ||
<input.dat> has lines with items: i j corr [ncorr] | <input.dat> has lines with items: i j corr [ncorr] | ||
-b option: <input.dat> is a binary file (4 bytes for each item) | -b option: <input.dat> is a binary file (4 bytes for each item) | ||
-w option: calculate weights from | -w option: calculate weights from number of correlated items (4th item on input line) | ||
-z option: use Fisher z-transformation | -z option: use Fisher z-transformation | ||
-f option: skip some calculations (fast) | -f option: skip some calculations (fast) | ||
| Line 32: | Line 38: | ||
* the number of vectors must be > 2*(low dimension). Typical number of dimensions is 2 or 3, but depending on the problem it could of course be much more. | * the number of vectors must be > 2*(low dimension). Typical number of dimensions is 2 or 3, but depending on the problem it could of course be much more. | ||
Python code is available [https:// | Python code is available as [https://wiki.uni-konstanz.de/pub/cc_analysis.py] under GPL. | ||
== Example == | == Example == | ||
Suppose we have 5 objects (e.g. images or crystallographic data sets), and measure their correlation coefficients against each other. These must be written to a file (named cc.dat in this example), and given as input to cc_analysis: | |||
<pre> | <pre> | ||
bash-4.2$ cat cc.dat | bash-4.2$ cat cc.dat | ||
| Line 48: | Line 55: | ||
4 5 0.8432 | 4 5 0.8432 | ||
</pre> | </pre> | ||
Please note that the CCs are rounded to 4 valid digits. This introduces a bit of noise. | Please note that the CCs are rounded to 4 valid digits. This introduces a bit of noise. In addition, the solution in this example is not overdetermined; 5 is just the minimum number of objects that can be represented in 2-dimensional space when given 10 unique correlation coefficients. | ||
<pre> | <pre> | ||
bash-4.2$ cc_analysis -dim 2 cc.dat solution.dat | bash-4.2$ cc_analysis -dim 2 cc.dat solution.dat | ||
CC_ANALYSIS version 29.10.2018 (K. Diederichs). No redistribution please! | CC_ANALYSIS version 29.10.2018 (K. Diederichs). No redistribution please! | ||
Linux turn31.biologie.uni-konstanz.de 4.18.14-1.el7.elrepo.x86_64 #1 SMP Sat Oct 13 10:29:59 EDT 2018 x86_64 x86_64 x86_64 GNU/Linux | Linux turn31.biologie.uni-konstanz.de 4.18.14-1.el7.elrepo.x86_64 #1 SMP Sat Oct 13 10:29:59 EDT 2018 x86_64 x86_64 x86_64 GNU/Linux | ||
| Line 99: | Line 104: | ||
The Python code produces the following output: | The Python code produces the following output: | ||
<pre> | <pre> | ||
python /usr/local/bin/cc_analysis.py 2 cc.dat | bash-4.2$ python /usr/local/bin/cc_analysis.py 2 cc.dat | ||
=== | === | ||
Correlation matrix parsed from infile: | Correlation matrix parsed from infile: | ||
| Line 158: | Line 163: | ||
Finished outputting 2-dimensional representative vectors! =) | Finished outputting 2-dimensional representative vectors! =) | ||
</pre> | </pre> | ||
The 5 lines at the bottom give the solution. The coordinates agree with those of the Fortran program within a rms deviation of 0.0055 | The 5 lines at the bottom give the solution. The coordinates agree with those of the Fortran program within a rms deviation of 0.0055 after mirroring across the x axis (inverted solution) and rotating by half a degree. | ||