Machine Learning Meetup Notes: 2010-06-30

Mike's bio overview:

amino acids build proteins

20 amino acids

• protein has an amino acid sequence (three bases make up an amino acid)
• dna comprised of 4 bases: A, T, C, G
• rna comprised of 4 bases, A, U, C, G
• A goes with T
• C with G

every three bases is a codon, dave wrote a script that will take the codons and map them to their amino acids

• protease - are a type of proteins that cleave other proteins?
• reverse transcriptase - takes viral rna and transcribes it into dna
• sends mrna (bad) into the ribosomes
• they replicate very fast in your immune cells and thats how they kill them

99 amino acids in protease (297 dna bases)

reverse transcriptase is not predictable - each sequence is a different length

Possible Features:

• for the OR acids, perhaps create all possible combinations and weight them by 1/(number of combinations), normal rows weight = 1
• find most probable sequences (T)
• correlating permutations (T)
• molecular weight/length (E/Th)
• acidity/charge
• edit distance (differences between the sequences), use to cluster (A)
• list of known resistant mvt sites (M)
• find out which sites are most variable

Liebenstein?

for each site, and look at frequency of each amino acid

could put into a tree classifier