LRUG__Lightning_Talk__11th_Feb_2013.pdf

Transcript

1. Afra collectively annotating genomes
2. Background • how does life work • why don't naked mole rats ever get cancer ◦ but humans do • can we contain the population of mosquitoes without the side-effect of insecticides The answers lies in genomes.
3. Genomes • stored in each cell of an organism ◦ in the form of DNA ◦ represented as sequence of 'A', 'G', 'C', and 'T' • parts of genomes are "genes" ◦ encoded instructions ◦ define the shape, size, behavior, lifespan and disease susceptibility Obtaining a genome sequence is easy. But finding genes is difficult. (wizz past in 20? or split in two?)
4. Gene on a DNA strand gene model
5. Gene finding • genome is run through prediction software ◦ guess the location of gene ◦ return an approximate model • predicted gene models are not accurate ◦ incorrect feature boundaries ◦ miss genes altogether ◦ combine two genes or split one
6. Refining gene models • experts go through the predicted gene • compare it to the evidences • manually refine the gene model
7. It's difficult • we need lots of humans to manually go through the predicted gene models Crowdsource!
8. How? • present gene models to users ◦ easier to curate models to beginners ◦ harder ones to experts • attempt to validate submissions automatically • when in doubt reach out to a more experienced user
9. gene_model.difficulty • easier to find mistakes in prediction with ◦ multiple supporting evidences ◦ "good" evidences • we look for these and assign a gene model a difficulty score
10. automatic validation >> candidates = gene_model.candidates [gene_model1, gene_model2] >> if refinements.first == refinements.last # commit else gene_model.difficulty.bump end
11. • some edits are easier to validate ◦ exon-intron boundaries ◦ EST alignment • validate these on the client side ◦ provide immediate feedback ◦ compute putative score
12. • present gene models to users ◦ based on their interest Tag genes before-hand. On signup • ask users to pick tags of interest to him from a tag cloud {needs to be worked on}
14. TaskScheduler >> user = session.user >> difficulty_level = linear_map user.score >> tasks.first(:difficulty => difficulty_level)
15. task - states - submissions - gene
16. ◦ based on user's experience level ◦ • collect their refinements as a possible candidate • validate user's submission • bump their score
17. Lifecycle of a gene model - this is the part where different agents come together to refine a gene model {include FSM diagram}
18. Client side • RIA • /tasks/next
20. Thank you :) -- yeban