SJSU ML Club's Ribonanza Project

SJSU ML Club's attempt at the Kaggle Ribonanza project hosted by Stanford.

To run our project, simply run python main.py.

About the project

• We used a continuous transformer architecture for our model. After comparisons with LSTM, Convolutional, and plain Feedforward models, the transformer architecture performed the best. We used a variety of techniques to make and improve our model.
• We used base-pair probability matrices to boost our model's performance. Base-pair probability matrices represent the probability that a certain base in an RNA strand will be paired with another base. Thus, it makes sense that this would help our model predict an RNA strand's 3D structure
• We used pseudo-labeling to train on the test dataset, boosting our leaderboard score by around 0.002

Members

• Haydon Behl
• Neal Chandra
• Eliot Hall

Score

Our best score was slightly less than 0.15