I have some questions about ResPRE

This forum is shown on the index page along with all topics.

Moderator: robpearc

cuilei
Posts: 7
Joined: Sun May 23, 2021 4:56 am

I have some questions about ResPRE

Post by cuilei »

Dear zhangyang laboratory:
Hello! I am a student who is new to bioinformatics, and I am studying the thesis of ResPRE recently. ResPRE uses the coupling precision matrix to achieve high-accuracy protein contact prediction. It has achieved excellent results. Is an outstanding work.I sincerely hope that some questions about this paper can be answered. :D :D :D
Question 1:
The model in ResPRE is trained through the Pytorch framework. So in the process of using Pytorch to train, how are the two functions __getitem__() and __len__() implemented in the DataLoader class when loading the data set? Can you provide the source code to learn?
Question 2:
In the training process, when loading data, is a sequence as input data for training or residue pairs as input data?
Attachments
2224.png
2224.png (231.56 KiB) Viewed 4146 times
115.png
115.png (17.9 KiB) Viewed 4146 times
robpearc
Posts: 43
Joined: Sun May 02, 2021 5:49 pm

Re: I have some questions about ResPRE

Post by robpearc »

Hello,

Thank you for your interest in ResPRE!

1) The len() function will internally call __len__ to return the length of the object.

Our implementation is:
def __len__(self):
return self.length

The __getitem__ function will vary depending on your data loader implementation. We have not released our ResPRE training script; however there should be many examples online that you can find to build your own implementation of a data loader. Essentially, the function will load the true contact map and any precomputed features such as the precision matrices and Potts Model for a training protein, which are stored in a python dictionary. The function returns the dictionary of features and ground truth contact map.

2) The sequence is used along with other features, including the pairwise features such as the Potts model and precision matrices.
cuilei
Posts: 7
Joined: Sun May 23, 2021 4:56 am

Re: I have some questions about ResPRE

Post by cuilei »

Sincerely thank you for your detailed answer. Thank you very much! :D :D :D
In addition, I have another question.
Q:Does the process of training ResPRE use supervised learning or unsupervised learning? If it is supervised learning, the sequence needs to be labeled before training. So what is this label?
liyangum
Posts: 8
Joined: Tue May 11, 2021 1:42 am

Re: I have some questions about ResPRE

Post by liyangum »

We used supervised machine learning to train ResPRE models for contact map predictioon. The supervised information should be the contact maps (binary matrix, 1 for being in contact) from pdb structures.
cuilei
Posts: 7
Joined: Sun May 23, 2021 4:56 am

Re: I have some questions about ResPRE

Post by cuilei »

Thank you very much! I figured out the problem.:D :D :D
niamul21
Posts: 1
Joined: Mon Jun 07, 2021 3:48 am

Re: I have some questions about ResPRE

Post by niamul21 »

robpearc wrote: Tue May 25, 2021 11:48 am Hello,

Thank you for your interest in ResPRE!

1) The len() function will internally call __len__ to return the length of the object.

Our implementation is:
def __len__(self):
return self.length

The __getitem__ function will vary depending on your data loader implementation. We have not released our ResPRE training script; however there should be many examples online that you can find to build your own implementation of a data loader. Essentially, the function will load the true contact map and any precomputed features such as the precision matrices and Potts Model for a training protein, which are stored in a python dictionary. The function returns the dictionary of features and ground truth contact map.

2) The sequence is used along with other features, including the pairwise features such as the Potts model and precision matrices.
The answer seems very much helpful for me as well. Thanks a million.
Post Reply