Presentations

Bioinformatics

• Genetic sequences
• Letters are used to represent molecules like amino acids or nucleic acids.
• These monomers assemble into long chains.
• There can be very long range dependencies.
• Splitting into “sentences” (e.g. genes, protein domains) is non-trivial.

ML on Biological Sequences

• Character level language models
• Downstream tasks include:
  • Sequence or character classification e.g. function prediction
  • Sequence or character regression e.g. solubility prediction
  • Sequence alignment
  • Structure prediction
  • Functional network prediction
• Usually uses bi-directional transformers.

Tokenization