BioModel Discussion Issue for eos4cxk

[Zainab-ik]

Summary !!!

• BioModel Name: Zeng2022 - Prediction of molecular properties and drug targets using a self-supervised learning framework

• BioModel Tag: fair AIML, Machine learning model, Ersilia

• Metadata Comments:

  1. The model performs both classification and regression task; I should focus only on the classification, right?
  2. The paper mentioned the model predicts molecular properties and drug target of a compound. Is the drug target prediction feature implemented?
  3. Also, I noticed the repository majorly mentioned COVID assay; the implementation of the molecular property prediction is tailored towards SARS-COV assays alone, right?
  4. Do i go ahead and list all the 13 COVID assays, or its classification or both?
  5. This is a multiple output model; for Ersilia implementation, what is the specific output; prediction of anti-SARS-CoV-2 agents or the more general output like; drug-like property assessment (brain permeability, drug metabolism and toxicity)?.
  6. The checkpoints are for COVID assays, the output would be antiviral prediction alone, right?
  7. Does active and inactive fit into this model scope; it seems to me like a computer vision model and there was no mention of data binarization?

• Contributor:

  roles	Names
  Curator	@Zainab-ik
  Code contributor	@DhanshreeA
  Reviewer	@GemmaTuron

• Curation status : manually curated

• Annotation file: here

[Zainab-ik]

@GemmaTuron This model is ready for review.

[GemmaTuron]

Hi @Zainab-ik :

The model performs both classification and regression task; I should focus only on the classification, right? As you can read in the model description and interpretation we are implementing 13 classifiers The paper mentioned the model predicts molecular properties and drug target of a compound. Is the drug target prediction feature implemented? Likewise, the interpretation field is showing which activities are being predicted - please make sure those are reflected on the metadata. Some assays refer to specific protein inhibitions, shouldn't these be on the metadata? Also, I noticed the repository majorly mentioned COVID assay; the implementation of the molecular property prediction is tailored towards SARS-COV assays alone, right? I am not sure to which you are refering as molecular property out of the 13 predictors implemented Do i go ahead and list all the 13 COVID assays, or its classification or both? what do you mean its classification? Please list the targets as defined in the model descripttions This is a multiple output model; for Ersilia implementation, what is the specific output; prediction of anti-SARS-CoV-2 agents or the more general output like; drug-like property assessment (brain permeability, drug metabolism and toxicity)?. The outputs are several, including prediction fo activity for sars-cov2 properties but also toxicity in human cells for example. Please complete the metadata accordingly The checkpoints are for COVID assays, the output would be antiviral prediction alone, right? There is other more general predictions like toxicity Does active and inactive fit into this model scope; it seems to me like a computer vision model and there was no mention of data binarization? The pre-trained model is ImageMol, the fact that it represents molecules as images has nothing to do with the output of the model (active, inactive)

Make sure to read about the 13 assays we are implementing and complete the metadata accordingly

[Zainab-ik]

Metadata completed accordingly.

[GemmaTuron]

Hi @Zainab-ik

The field experimental method, stands for validation or training data? As we discussed we should only list it for validation The model also lists some activities for SARS Cov1 maybe? please double check

[Zainab-ik]

Training data, and removed. I missed it, SARS-Cov is SARS-Cov1, noted. I added the activity.