Collaboration with James's team

[yk-tanigawa]

Decided to keep track on the progress on the collaboration project with James's team on GitHub as well

meetings

2017/1/11

• Presented my project to their team
• https://github.com/rivas-lab/presentations/blob/master/yosuke/20170111_yosuke.pdf

2017/1/23

• Journal club & discussion
• I presented Genome Graph pre-print paper
  • https://github.com/rivas-lab/presentations/blob/master/yosuke/20170123_yosuke.pdf
• They presented Deep Variant

2017/1/30

• discussion
  • Haplotype inference (& phasing) with deep net ??

2017/2/6

• discussion
• we will initially focus on chromosome 20
• Yosuke have pointed out GIAB data set and provided download script for James's team so that they can prepare 'label' for their variant caller
• They also have bam file of selected long reads from Nanopore consortium consortium data, which will be used to construct pile-up image (input) for their deep network
• Yosuke will continue to work on my project to produce posterior probability of haplotype given reads

[yk-tanigawa]

2017/2/13

presentation

• Yosuke presented haplotype inference estimate
  • https://github.com/rivas-lab/presentations/blob/bf5eac8c08e0537d2a889334b2e67483ed557047/yosuke/20170213_yosuke.pdf
• Ziyi presented their progress to produce pile-up image
  • 401 x 100 image
  • red: type of base
  • green: base call quality
  • blue: strand (+ / -)
  • alpha: read vs. ref / read vs. call

discussion

• read specific error rate
• model general likelihood model
  • Pr(data | hap) = f(read, signal, base call Q, and more | haplotype ; theta)
  • where, theta is a parameter we would like to optimize
  • training step
  • argmax_\theta Pr(correct haplotype | data; theta)

    Next step

• James's team will apply their method to larger data set (>= 12.5 kb read length)
• Try to incorporate signal data (raw data in fast5 file)

[yk-tanigawa]

• new bam file has been prepared.
  • summary stats are written in #15

[yk-tanigawa]

Their progress:

• They expanded the window size to 1kbp
• Sample image of chr20:103,017-104,017
• The same region on iGV
• asking to start training step if they're ready to do so

[yk-tanigawa]

Clarification of their prediction task (discussion on slack)

• Asking them to implement a neural network, where
  • input: pileup image
  • output: likelihood of phased genotypes

[yk-tanigawa]

Agenda for meeting at 3/13

Here is the agenda for tomorrow:
- Read specific error rate (Yosuke)
- CNN approach for Base specific error rate ( @chen , @ziyi )
- Feature importance analysis ( @chen  )
- Yield plot analysis (if time allows, Yosuke)

Let me know if there any other interesting topic you’d like to discuss.

I also think it would be great if we can evaluate the performances of three models (fixed-global, read-specific, base-specific) error rates.
@ziyi , @chen : Is there any chance that you can share the base-specific error rate so that I can use that for evaluation?