Inspiration
When asked about we would tackle this project, we decided, considenring how sensitive the predicted output is, that we could use a democratic vote among different models to make a precise guess. This is inspired from nuclear powerplants that use 4 separate computers (AI) to decide on site-critical actions by voting among themselves.
What it does
We produce a csv file containing predictions about the toxicity (hERP inhibator or not) of molecules with given smiles and fingerprints.
How we built it
We use three models : _ Random Forest _ , _ SVM (Support Vector Machine _ and _ Deep Neural network _ . Their predictions are then compared to implement the vote. The models are traine on a supplied set of data.
Challenges we ran into
We had alot to troubleshoot because the output data of each model was different and had to be reshaped to then be used in the vote system.
Accomplishments that we're proud of
it's working... more or less.
What we learned
We discovered a lot about the different AI models and their respective strenghths and weaknesses.
What's next for hERp-ES Team
Holidays or application to other cases maybe ?
Built With
- python
- scykitlearn
- tensorflow
Log in or sign up for Devpost to join the conversation.