soedinglab / bammmotif2 Goto Github PK

There are a two issues with the homogenous markov models stored in the hbp and hbpc files

• the precision is only four digits, which makes the probabilities not sum to 1
• the joint probabilities in the hbp file sum to a value greater than one. It seems that the first entry of orders >1 is way too large.

Server does not stop updating when job crashes, in case of Motif-Motif comparison.

Hi, I have been using the server version for some time for motif discovery and I wanted to add the shell version to my pipeline. However, when I run:
BaMMmotif DIRPATH FILEPATH

I get the following message:

Error: No initial model is provided.

Am I missing some necessary step/required file?

We should check for availability of boost in the cmake file so that we can output a proper error message during build phase.

Failure if -n is larger than the total number of motifs in the pwm file.
It is better to use -n as a kind of maximum of pwm's you want to process...

If I am not mistaken, we currently do not have specifications for the some of our own file formats such as occurrence files.

Similar to MMseqs, I think we should start using the wiki for expanding the documentation step by step.

Is there a rationale behind not defining the information content as max_entropy - entropy here?

https://github.com/soedinglab/bamm-private/blob/master/R/plotBaMM.R#L259-L261

informationContent <- function( x, base=2 ){
    ifelse( all( x > 0 ), 2 + sum( x * log( x, base ) ), 0 )
}

This sets the information to zero if there's at least one nucleotide occurring with rel. frequency of zero, leading to very odd behavior:

> informationContent(c(0.25, 0.25, 0.25, 0.25))
0
> informationContent( c(0.01, 0.01, 0.01, 0.97))
1.75805926714679
> informationContent( c(0.0001, 0.0001, 0.0001, 0.9997))
1.99558094270164
> informationContent(c(0, 0, 0, 1))
0

The conditional probabilities in BaMM model/background files are sometimes larger than 1, by a margin that is higher than numerical instability would suggest: e.g. 1.12 in the attached model file, and 3.09 in the attached background model.

varlen_seqs.hbcp.txt
varlen_seqs_motif_1.ihbcp.txt

command call:
BaMMmotif ./mcf7_GATA_narrow /home/mmeier/git/PEnG-motif/scripts/mcf7_GATA_narrow/mcf7_GATA_narrow.fasta --PWMFile /home/mmeier/git/PEnG-motif/scripts/mcf7_GATA_narrow/mcf7_GATA_narrow.tmp.out --FDR --savePvalues -K 2 --zoops

Result:
terminate called after throwing an instance of 'std::length_error'
what(): vector::_M_default_append

Backtrace with gdb:
#10 Motif::initFromPWM (this=0x51546f0, PWM=PWM@entry=0x66b680, asize=4, count=) at /home/mmeier/git/bamm-private/src/bamm/Motif.cpp:255

when printing LW1:
LW1 = -7

There seems to be a very short sequence in the dataset:

chr8:93080412-93080414
AC

Can you catch this?

current build failure on my mac

/Users/ch/repo/bamm-private/src/shared/SequenceSet.cpp:357:50: error: cannot take the address of
 an rvalue of type 'std::__1::basic_ostringstream<char, std::__1::char_traits<char>, std::__1::a
llocator<char> >'
                                                header = static_cast<std::ostringstream*>( &( st
d::ostringstream() << ( N+1 ) ) )->str();
                                                                                           ^  ~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

http://pastebin.com/AfXfdZwH

We will need to call the R scripts from various other scripts. It'd be best to install them automatically in the cmake file. It should be configured such that make install moves the script to the bin folder of the given installation prefix.

For this to work properly all R files should use /usr/bin/env Rscriptin the shebang and avoid hardcoding the path to Rscript

subsample input sequences when input set is too large?

• skip N's while reading in sequences;
• adjust occur2bed.py so that it automatically converts occurrence file to bed file when header information in input FASTA file is provided;
• read in multiple BaMM files at the same time;
• look into the problem where it takes longer time for EM optimization as given BaMM than MEME as seeds;
• look into the problem where log odds distribution of positives is shifted to the left of that of negatives;
• investigate on why TGACTCA motif is everywhere;
• write out the hyper-parameters to the BaMM model file: e.g. k, alpha
• check if the bed file positions are correct.
• how to treat the short sequences fairly.
• v[k] does not sum up to Y[k+1]
• ignore the usual end-of-line char in the input files.
• write Uni_test for tracing the changes.