Does anyone have experience with QSAR? I looking for examples of workflow. Unfortunately google (knime workflow qsar) is not enough ;) Does anyone know of tutorials or examples of workflow?
Thx
Hi,
not sure precisely what you are after. If you want to build a QSAR model for predicting a value such as activity on a new molecule based on existing data then try out these nodes.
Tree ensemble learner regression and tree ensemble predictor regression nodes if you want predict an exact numerical value.
or
Fingerprint Bayesian learner and predictor nodes if you want to predict a class rather than an exact value, ie very active, moderately active, poorly active, inactive.
To use these you will need to generate molecule fingerprints first. Use the rdkit fingerprint node to do this and select Morgan fingerprint which will be a good one for this purpose. Now Feed in these molecules with their fingerprints and with known activity data to the learner node and new molecules with no measured data to the predictor node.
Of course in reality it's more complicated than just the above as you need to validate the accuracy of the model by taking molecules with known measured data and then predicting values on these with the predictor node. (Use row sampling node for this) You can generate an R2 to assess accuracy and use the scorer node to assess binning accuracies. This should give you an idea if your QSAR model is predictive.
hope this helps
simon.
1 Like
Hi Simon,
thanks for your answer.
My experience with QSAR is limited to Matlab and few descriptors.
When I finding an error in the calculations I have to start again and again...
I want to automate the process and expand range of models and descriptors.
I think the best way study is the study by analyzing the finished solution :)
That's why I'm looking for examples of workflow with descriptors (and selection of the best descriptors), models (eg. PLS, MLR, kNN, FFNN), analyzing (eg. r2, q2, RMSE, cross-validation), and visualizations.
eg basic examples:
http://www.modena-cost.eu/Portals/3/events/Dave%20Winkler%20Training%20Material%20ROME.zip
@richards99, I am a computational chemist who is new to KNIME, and your post was extremely informative. Thank you!
1 Like