Neural Network? Workflow for >6000 compounds by KNIME

Finally had a time to play with KNIME keras integration. But didn't know how to use them at all so took a parts from 09_Wide_and_Deep_Learning_on_Census_Dataset which is up on EXAMPLES. So here it is

I calculated ECFP and MACCS for fun this time. Then changed them to one-hot vector for keras input purpose. Then copied metanode from EXAMPLE 09 with bit of deletion coz some nodes were unnecessary in my case. Finally dense them to 512 and then 2 for a very simple NN. I think you can add/change layers to Deep Learning ish flow too.

I hope I'm doing this right... I have not much confidence but score says 0.774. Not a bad start considering SVM was something like 0.79ish??? I should parameter optimze this work flow as well. Wanna play around with other KNIME keras DL nodes too.

This is ver fun so I might update this workflow but for now I also played with SHAP so wanna blog that too. KNIME is so so so easy. I can just pass this to my colleague and they can check for me. Pythoning was very nice too so I might use Python nodes as well...