Last October, at D4 USA, Gerald Sun, Senior Data Scientist, AstraZeneca, demonstrated how to get more out of convolutional neural networks to solve genomics problems in drug development. In his presentation Gerald also shared some graph examples to explain how they can serve as tools for general-purpose data representation for machine learning.
Key points:
- Why CNNs are the most prolific data representation tool
- Why CNNs and optimised data representations work for predicting tumour purity
- “If the representation enables you to interpret your data, it’ll probably enable the model”
The publication mentioned in the presentation can be found here