Inspiration
We are both heavily interested in biochemistry, but haven't worked with machine learning too much until this challenge. It gave us a chance to experiment with some standard data science and machine learning tools.
What it does
Our model predicts 'drug binding' or 'non-drug binding' for any query residue on an AlphaFold2 predicted protein model.
How we built it
This was our first time applying machine learning to a dataset of this scale, and we built it with TensorFlow and Keras. We figured a neural network would be best given the large amount of data and features, and attempted hyper-parameter optimized to the best of our ability.
Challenges we ran into
The main issues we came across were cleaning up the data, and figuring out which features were of most importance on the chemistry side of things.
Accomplishments that we're proud of
We were proud to make a working neural network and learn the fundamentals of TensorFlow and Keras.
What we learned
We learned the basics of how to implement a neural network with TensorFlow and Keras, as well as some common techniques for optimization.
What's next for Cyclica AlphaFold2 Challenge
We will continue to develop our model over the coming weeks, and our goal is to optimize our neural network to the best of our ability, and continue learning how to best apply machine learning techniques and when.
Log in or sign up for Devpost to join the conversation.