search

isjakewong / MIRACLE

Multi-view Graph Contrastive Representation Learning for Drug-drug Interaction Prediction
https://arxiv.org/abs/2010.11711
42 stars 8 forks source link

have the implementation done correctly? #4

Closed Namkyeong closed 2 years ago

Namkyeong commented 2 years ago

스크린샷 2022-09-06 오후 12 12 38

Hello authors thank you for your great work predicting drug-drug interactions!

I have questions on the implementation. Where can I find the interaction prediction in the code? It seems like the model simply predict the interaction between drug i and j using only feature of i.

Thank you! Namkyeong

isjakewong commented 2 years ago

Hi, in this final version, we directly use a FFN to predict the interaction score. The reason why we simplify this operation is that drug synergy information has been flowing into each other in previous message-aggregating and further contrastive learning. Thus, we can easily apply a FFN to disentangle the hidden patterns for prediction. Maybe you could try cross-attention to improve the fusion of different drug representations.