After viewing recent advances in predicting adverse drug reactions along with those in 3D small molecule alignment, we were inspired to attempt to couple the two.
What it does
We leverage 3D structure and gene expression data to show improvements in accuracy and significant improvements in training speed for adverse drug reaction prediction using extremely randomized trees.
How we built it
We built it in python, using scikit-learn, several packages including multiprocessing and LS-Align.
Challenges we ran into
We had difficulty with the general time requirements for certain steps. We couldn't use flexible alignment as our devices lacked sufficient processing power. Additionally, we were unfamiliar with working with molecular datasets.
Accomplishments that we're proud of
The general speed up we achieved as well as our integration of multiprocessing to speed up the overall program.
What we learned
We learned that even in a significantly reduced form, 3D data has the potential to significantly impact ADR predictions.
What's next for Prediction of ADRs using 3D Data, GE data, and Extra Trees
We hope that more research can be done on the topic of including 3D data.