Deep Learning for Synthetic Biology

03 Jan 2022 in Projects

Summary of research progress on Synthetic Biology powered by Deep Learning.

Preliminaries About Synthetic Biology

• Glossaries:
  • Protein
  • RNA (tRNA, mRNA, rRNA)
  • Gene
  • DNA
• Protein synthesis pipeline: DNA (including coding regions, a.k.a. genes, and non-coding regions) -> RNAs -> Proteins
• Orthogonal translation systems
• DNA assembly

Deep learning techniques

• Biological senses: Biological problems to solve, what factors need to be considered.

• Data collection, curation, pre-analysis before modelling
  • active learning for informed training data building
  • human-in-the-loop data curation, cleaning, and model improvement
• DL models:
  • transformers,
  • perceivers,
  • GNN, etc
• Adverse/Challenging cases/scenes/settings:
  • meta-learning (small data for each separate task),
  • few-shot setting
  • NAS,
  • label noise,
  • sample imbalance, etc
• Model interpretation, analysis, and further improvement
  • Explainable or interpretive or transparent AI
  • Uncertainty measurement and explanation
• General DL aspects:
  • memory efficient
  • fast inference
  • optimisation techniques
  • loss functions
  • generalisation: avoid overfitting and underfitting