[DL Edition] T038: Protein Ligand Interaction Prediction

[Old-Shatterhand]

Description

Proof of concept for GNN-based protein ligand interaction prediction in talktorial T038

Todos

• Find an appropriate dataset
• Fix bug in code to be able to train on batches larger than one sample

Questions

None

Status

Initial draft for further discussion

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[Old-Shatterhand]

Pull request update

After some minor updates on the nodebook, the following work is left.

TODOS:

• Find an appropriate dataset for training
• Discussion: Put plots and explanations of training based on selected dataset
• Questions: Come up with questions about the content

[Old-Shatterhand]

Talktorial review

Pullrequest of talktorial about GNN-based protein-ligand interaction prediction.

Details

• Talktorial ID: T038
• Title: Protein-Ligand Interaction Prediction
• Original authors: Roman Joeres
• Reviewer(s): Andrea Volkamer, tbd
• Date of review: DD-MM-YYYY (tbd)

Content

• One line summary: Introduction of GNN-based protein-ligand interaction prediction
• Potential labels or categories (e.g. machine learning, small molecules, online APIs): Kinases, Machine Learning, Graph Neural Networks
• Time it took to execute (approx.): 1 hour
• I have used the talktorial template and followed the content and formatting suggestions there
• Packages must be open-sourced and should be installable from conda-forge. If you are adding new packages to the TeachOpenCADD environment, please check if already installed packages can perform the same functionality and if not leave a sentence explaining why the new addition is needed. If the new package is not on conda-forge, please list them and their intended usage here.
  • biotite, pypdb, chembl-webresource-client, rdkit: Already in TeachOpenCADD
  • torch 1.10.1, torch-geometric 2.2.0, torch-cluster 1.6.0, torch-scatter 2.0.9, torch-sparse 0.6.13, torch-spline-conv 1.2.1, cpuonly 2.0: From "dubious" sources (either own conda-channel (pyorch) or installed as pip wheel. All of them only for cpu
• Data must be publicly available, preferably accessible via a webserver or downloadable via a URL. Please list the data resources that you use and how to access them:
  • KiBA dataset: Access via website

Content style

• Talktorial includes cross-references to other talktorials if applicable
• The table of contents reflects the talktorial story-line; order of #, ##, ### headers is correct
• URLs are linked with meaningful words, instead of pasting the URL directly or linking words like here.
• I have spell-checked the notebook
• Images have enough resolution to be rendered with quality, without being too heavy.
• All figures have a description
• Markdown cell content is still in-line with code cell output (whenever results are discussed)
• I have checked that cell outputs are not incredibly long (this applies also to DataFrames)
• Formatting looks correctly on the Sphinx render (bold, italics, figure placing)

Code style

• Variable and function names follow snake case rules (e.g. a_variable_name vs aVariableName)
• Spacing follows PEP8 (run Black on the code cells if needed)
• Code line are under 99 characters each (run black-nb -l 99)
• Comments are useful and well placed
• There are no unpythonic idioms like for i in range(len(list)) (see slides)
• All 3rd party dependencies are listed at the top of the notebook
• I have marked all code cell with output referenced in markdown cells with the label # NBVAL_CHECK_OUTPUT
• I have identified potential candidates for a code refactor / useful functions
• All import ... lines are at the top (practice part) cell, ordered by standard library / 3rd party packages / our own (teachopencadd.*)
• I have used absolute paths instead of relative paths

  HERE = Path(_dh[-1])
  DATA = HERE / "data"

Website

We present our talktorials on our TeachOpenCADD website (https://projects.volkamerlab.org/teachopencadd/), so we have to check as well if the Jupyter notebook renders nicely there.

• If this PR adds a new talktorial, please follow these steps:
  • Add your talktorial to the complete list of talktorials here (at the end).
  • Add your talktorial to one or multiple of the collections here. Or propose a new collection section in your PR.
  • Add your talktorial's nblink file by running python generate_nblinks.py from within the directory teachopencadd/docs/talktorials.
  • Please complile the website following the instructions here.
• Check the rendering of the talktorial of this PR.
• Is your talktorial listed in the talktorial list?
• Is your talktorial listed in the talktorial collections?
  • Add a picture for your talktorial in the collection view by following these instructions.

[gerritgr]

• I will (and elsewhere)-> We will? (I think in other notebooks, it is third person, have not checked though).
• "field of protein ligand interaction prediction" -> "protein-ligand"
• Maybe explain the terms protein and ligand very shortly in the intro.
• Titles are lowercase (Sentence Case) in the other notebooks.
• "..., I'll link to this otherwise, I'll explain new things below." -> "... , I will link to this. Otherwise, I will explain new things below."
• "one wants to" -> wants
• There are some other typos, but Grammarly or so can catch these I guess.
• simple Feed-forward Neural Network (FNN): do you mean MLP? GNNs are also technically feed-forward networks.
• State in the beginning of the workflow that this is a binary classification task.
• " from the PDB entry with ID 4O75." link to a talkturial explaining PDB earlier.
• explain C_alpha
• In the "Technical background", say that we use the same GNN architecture for both ligands and proteins.
• I think the BCE explanation should be clarified. What are the negative and positive samples here (binding and non-binding?)?
• suppress DtypeWarning and add a comment explaining what kiba_preprocessing is doing
• "Storing and representing data in PLI-prediction is a bit different from other neural networks. ": You mean the input data (i.e., the graphs), right? Maybe clarify this. Also, even though it is obvious, PLI was not introduced as an abbreviation.
• Maybe merge together with the "Data Points" subsection.
• I have not checked the code yet.

[Old-Shatterhand]

I implemented Gerrits comments and uploaded a new notebook.

[mbackenkoehler]

• Describe interaction a bit more. Is it only about binding? If so maybe mention classical approaches to the same problem.
• For example (missing comma)
• You could motivate the model with virtual screening of compound libraries, for example
• TBC in the end.

Reply via ReviewNB

[mbackenkoehler]

• Maybe change workflow to 'model' something similar; The workflow would include data prep, training and so on, I presume.
• Simplify the second sentence.
• "We will only use the information if an interaction exists or not," -> This sounds a bit strange. Just say right away, you are transforming the task to a classification
• introduction of FNN acronym is missing
• link to rcsb.org for 4O75

Reply via ReviewNB

[mbackenkoehler]

• remove 'in protein ligand interaction prediction' in second sentence
• typo in C_{alpha} (last sentence above fig.)
• fig 2 caption: 'protein structures as graphs' is maybe better; typo: representations
• missing figure 3?

Reply via ReviewNB

[mbackenkoehler]

• second sentence has 'use' twice
• third sentence: missing comma?

Reply via ReviewNB

[mbackenkoehler]

• clarify if epsilon is a (hyper-)parameter
• not sure if onehot or one hot
• last par: final element to finalize our -> final element to our GNN
• full stop after pooling function.
• "For simplicity reasons" sounds wrong. Maybe just "For simplicity, " or "For the sake of simplicity"

Reply via ReviewNB

[Old-Shatterhand]

Regarding second point: I guess, one-hot is the solution, Grammarly would suggest. I changed the text occurrences to "one-hot".

[mbackenkoehler]

Line #25.            with open(os.path.join(self.folder_name, "tables", "ligands.tsv"), "r") as data:

use path library for consistency

Reply via ReviewNB

[mbackenkoehler]

Line #32.                (filename[:-4], pdb_to_graph(os.path.join(os.path.join(self.folder_name, "proteins", filename)))) for filename in os.listdir(os.path.join(self.folder_name, "proteins"))

pathlib (see above)

Reply via ReviewNB

[mbackenkoehler]

Line #62.                # then split the data and store them for later reuse without running the preprocessing pipeline

• consider a torch data splitter?

Reply via ReviewNB

[Old-Shatterhand]

Would be too complex here, I'd say. Here, you easily see what's happening and how and why.

[mbackenkoehler]

Line #1.    class Encoder(torch.nn.Module):

Maybe use encoding in fig 1 for consistency

Reply via ReviewNB

[mbackenkoehler]

describe what the training does. I know, it's obvious, but maybe it helps. Make the reader recall the BCE loss and mention that Adam is a standard choice etc...

explain why there is a single epoch, maybe encourage the reader to try higher values.

Reply via ReviewNB