• Professor: Mingon Kang

interpretable deep learning

machine learning can learn new patterns from data

• deep learning can learn non-linear patterns from data
• in science, model interpretability is more important than accurate prediction.
• intrinsic interpretation is derived from the model’s construction.

pathway-informed deep learning

• gene data is used to inform the construction of the neural network
  • leads to better interpretation
  • activation of node indicates more activity in a specific pathway

evidential deep learning

• github: datax-lab/EPICK

• enzyme commission number classifies enzymes based on chemical reactions

• multi-label classification shows high number of false positives

• EPICK provides potential active sites which can be compared to well-known sites.

• integration of pathological data and genomic data allows for better classification.

• pathological image allow classification of genetic abnormalities