Hi,

I used cisgenome for most of our ChIP-Seq analysis. Recently we have Drosophila ChIP-Seq data, the computing FDR step using hts_windowsummaryv2 gave me all invalid neg binomal data as below.

# Window_Size=100 Poisson_Lambda=0.057520 Poisson_p=0.319700

NegBinom

ial_Alpha=0.029232 NegBinomial_Beta=-0.491803 NegBinomial_p=-nan

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 509299 0.301829 0.301829 1.000000 -nan

-nan

1 29295 0.017361 0.017361 1.000000 -nan

-nan

2 29665 0.017581 0.000499 0.028401 -nan

-nan

3 31363 0.018587 0.000010 0.000515 -nan

-nan

We also have control data, so I ran 2-sample analysis, and came with the similar result.

#Window_Size=100 dP0_hat=0.556655

#Poisson_Lambda=0.034540 Poisson_p=0.303430

#NegBinomial_Alpha=0.019155 NegBinomial_Beta=-0.445431

NegBinomial_p=-nan

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 494618 0.293129 0.293129 1.000000 -nan

-nan

1 17084 0.010125 0.010125 1.000000 -nan

-nan

2 15698 0.009303 0.000175 0.018795 -nan

-nan

3 15723 0.009318 0.000002 0.000216 -nan

-nan

Only when setting window_size beween 5-8, odd number shown on the last column.

#Forward+Reverse_Combined

# Window_Size=5 Poisson_Lambda=0.263733 Poisson_p=0.866340

NegBinomial_Alpha=0.373173 NegBinomial_Beta=0.414966

NegBinomial_p=1.000000

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 22458969 0.665504 0.665504 1.000000

0.632701 0.950709

1 5923167 0.175515 0.175515 1.000000 0.166864

0.950709

2 2874109 0.085166 0.023145 0.271760 0.080968

0.950709

3 1351820 0.040057 0.002035 0.050794 0.045266

1.130041

4 621905 0.018428 0.000134 0.007280 0.026978

1.463935

I skipped the FDR cutoff choose, and directly ran peakdetectorv2, cisgenome gave some reasonable result. I just don't know how reliable they are. I want to use strand specific window scan peak detection.

If any experts can point me to the hints, really appreciate.

litd

I used cisgenome for most of our ChIP-Seq analysis. Recently we have Drosophila ChIP-Seq data, the computing FDR step using hts_windowsummaryv2 gave me all invalid neg binomal data as below.

# Window_Size=100 Poisson_Lambda=0.057520 Poisson_p=0.319700

NegBinom

ial_Alpha=0.029232 NegBinomial_Beta=-0.491803 NegBinomial_p=-nan

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 509299 0.301829 0.301829 1.000000 -nan

-nan

1 29295 0.017361 0.017361 1.000000 -nan

-nan

2 29665 0.017581 0.000499 0.028401 -nan

-nan

3 31363 0.018587 0.000010 0.000515 -nan

-nan

We also have control data, so I ran 2-sample analysis, and came with the similar result.

#Window_Size=100 dP0_hat=0.556655

#Poisson_Lambda=0.034540 Poisson_p=0.303430

#NegBinomial_Alpha=0.019155 NegBinomial_Beta=-0.445431

NegBinomial_p=-nan

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 494618 0.293129 0.293129 1.000000 -nan

-nan

1 17084 0.010125 0.010125 1.000000 -nan

-nan

2 15698 0.009303 0.000175 0.018795 -nan

-nan

3 15723 0.009318 0.000002 0.000216 -nan

-nan

Only when setting window_size beween 5-8, odd number shown on the last column.

#Forward+Reverse_Combined

# Window_Size=5 Poisson_Lambda=0.263733 Poisson_p=0.866340

NegBinomial_Alpha=0.373173 NegBinomial_Beta=0.414966

NegBinomial_p=1.000000

#No_of_reads/window No_of_window percentage poisson_expected

poisson_exp/obs negbinomial_expected negbinomial_exp/obs

0 22458969 0.665504 0.665504 1.000000

0.632701 0.950709

1 5923167 0.175515 0.175515 1.000000 0.166864

0.950709

2 2874109 0.085166 0.023145 0.271760 0.080968

0.950709

3 1351820 0.040057 0.002035 0.050794 0.045266

1.130041

4 621905 0.018428 0.000134 0.007280 0.026978

1.463935

I skipped the FDR cutoff choose, and directly ran peakdetectorv2, cisgenome gave some reasonable result. I just don't know how reliable they are. I want to use strand specific window scan peak detection.

If any experts can point me to the hints, really appreciate.

litd

## Comment