hands-on protocol for contacts_to_distograms

[Yan-Yan-2020]

Hi,

Could you please share a hands-on protocol on how we can generate distogram with contact information? as a beginner, it seems hard for me to use the scripts (contacts_to_distograms.py) to build the distogram.

Thank you so much!
Yan

[lhatsk]

Hi Yan,

To generate distograms with the contacts_to_distograms.py script you need two inputs.

1. a CSV with your contacts where each row has the following format: ResiduePosition1 ResiduePosition2 FDR
   ResiduePositions are 1-indexed, i.e., starting from 1.
   FDR corresponds to the uncertainty for this particular contact, essentially the level of noise you expect. It's a number between 0 and 1. E.g., 0.05 would mean there is a 5% chance this contact is noisy/ wrong. In principle, you can use an FDR of 0 but this would be more akin to forcing constraints.
   Example:
   10 15 0.05
   30 50 0.2
2. the cutoff for your contact information in Angstrom (CA-CA).

The script currently only allows one cutoff. If you want to mix multiple cutoffs (which is possible), a quick hack would be to run this script with different cutoffs and concatenate the output files.

The output is another CSV with the distograms per contact pair that you can use for prediction.

Run like this:
python contacts_to_distograms.py --csv my_contact_file --cutoff 10 --output my_distogram_output

Let me know how it's working out with your data, I'm curious!

Kolja

[Yan-Yan-2020]

Hi Kolja,

I just tried the contacts_to_distograms.py script. it showed the following ERROR:

22 def get_uniform(cutoff, fdr):
23 d = np.ones(128)
---> 24 maximum_bin = np.argmax(distogram_bins > cutoff)
25 d[:maximum_bin] /= np.sum(d[:maximum_bin])
26 d[:maximum_bin] *= 1 - fdr
TypeError: '>' not supported between instances of 'numpy.ndarray' and 'str'

do you know how to fix it?

Thank you.
Yan

[lhatsk]

Sorry! It's fixed in the new version.

[Yan-Yan-2020]

Hi Kolja,
the new script works now.
Thank you so much!
Yan

[Yan-Yan-2020]

Hi Kolja,
i have a new problem on the 'preprocessing_distributions.py' script. what is the format of the infile "restraints.csv"? Could you please give an example for the csv file?

Thank you so much!
Yan

[Yan-Yan-2020]

specifically, it gave the errror:

cannot parse restraint type in line
('From', 'To', 'mu', 'sigma', 'type')

Thank you.
Yan

[grandrea]

hello,
if you run

preprocessing_distributions.py --help

it should give you the format of restraints.csv (I'll fix the way it's displayed and include an example and clarification in the readme)

The file should be made up of lines like (for example)

125,51,10.0,5.0,normal
16,10,10.0,5.0,log-normal

These lines specify

a restraint from resi 125 to resi 51 with a mean distance of 10 Angstrom, standard deviation of 5 and a normal distribution
a restraint from resi 16 to resi 10 with a mean distance of 10, standard deviation of 5 and a log-normal distribution

[Yan-Yan-2020]

yes, i did run the preprocessing_distributions.py --help it showed
"residueFrom, residueTo, mean_dist, stdev, distribution_type"

and i changed them to ('From', 'To', 'mu', 'sigma', 'type') in your original script, but it still do not work and showed cannot parse restraint type in line
('From', 'To', 'mu', 'sigma', 'type')

[grandrea]

could you attach your restraint.csv file here?

[grandrea]

(there should be no header in the file, just the comma-separated list of restraints with the lines like my comment above)

[Yan-Yan-2020]

it works now without headers. thank you so much!

[Yan-Yan-2020]

Hi,
Again, now I have questions on the predict_with_crosslinks.py script. In the READ.ME protocol, command shows
'python predict_with_crosslinks.py --distograms --checkpoint_path resources/AlphaLink_params/finetuning_model_5_ptm_CACA_10A.pt 7K3N_A.fasta restraints.csv uniref90.fasta mgy_clusters.fa pdb70/pdb70 pdb_mmcif/mmcif_files uniclust30_2018_08/uniclust30_2018_08'

i dissected into the following way:

python predict_with_crosslinks.py
--distograms restrains_distribution.csv (#csv file generated by preprocessing_distributions.py script)
--checkpoint_path resources/AlphaLink_params/finetuning_model_5_ptm_CACA_10A.pt
? 7K3N_A.fasta (#i guess it's the input file? then what script we should run?)
? restraints.csv (#why there is restraints.csv file here? what py script we should run?)
--uniref90_database_path uniref90.fasta
--mgnify_database_path mgy_clusters.fa
--pdb70_database_path pdb70/pdb70
? pdb_mmcif/mmcif_files (#what script we should run? like openfold command format? )
--uniclust30_database_path uniclust30_2018_08/uniclust30_2018_08

i left questions on each line of the commands.

Thank you so much!
Yan

[lhatsk]

Hi Yan,

--distograms is only a flag, it doesn't take an argument.
restraints.csv then corresponds to your restraints_distribution.csv.
7K3N_A.fasta -- you need to have some sort of FASTA input file with your sequence.
pdb_mmcif/mmcif_files -- yes, follow the OpenFold instructions, also for setting up the databases like uniref90, mgnify, and pdb70.

Kolja

[Yan-Yan-2020]

Thank you so much Kolja.

[Yan-Yan-2020]

Hi, again the error as following:

"RuntimeError: Error(s) in loading state_dict for AlphaFold:
size mismatch for xl_embedder.linear.weight: copying a param with shape torch.Size([128, 1]) from checkpoint, the shape in current model is torch.Size([128, 128])"

How to easily figure it out?
Thanks
Yan