Prediction of known human-protein and viral-protein interactions encountered issues

[lnyawen]

Hi Dan,

You have done an impressive job! I am very interested in this. I would like to use this tool to predict interactions between proteins of interest. Specifically, human NTCP has been repeatedly reported as a receptor protein for the hepatitis B virus via the preS protein. However, when I used your tool to predict the interaction between the two, the result was extremely low, with a value of 0.00023661248269490898. The model I used was danliu1226/PLM-interact-650M-VH.

How can I improve the current results?

I look forward to your response. Thank you!
Yawen

[liudan111]

Hi Yawen, thanks for your questions. The PLM-interact-650M-VH model is trained on a benchmark virus-human PPI dataset from 2019. Currently, we are working on creating a larger virus-human PPI dataset sourced from seven public databases, and we are training a new, more general virus-human model. This new model will be available soon.
If you are looking to validate our model before using it for virus-human PPI prediction, it is recommended to predict a larger set of virus-human PPIs for validation—such as 1,000 pairs. This will provide a more comprehensive evaluation of the model's performance. Thank you for your attention.