Second Eval

1 minute read


Second evalutation!

Hi there.

Several major events and exchange of ideas rolled around the corner. As a consequence, I made one significant change to my original project proposal, briefly summarized below.

  1. To keep up with the frontier of big-data genomics, I will be investigating how to stream binary files via memory-mapping as my second major deliverable. Douglas Bates very coincidentally initiated BEDFiles.jl just a week ago, which became the final straw to push me through with this drastic change of plans. I will possibly be spending all my time before JuliaCon 2018 on this part of the project.

  2. My original second deliverable - prior weighting of SNPs - have been passed down as a 6-weeks long summer project for 2 students participating in the BIG summer program. I will be overseeing these summer interns as opposed to being the main contributor for this part of the IHT project.

Summary of progress so far

I completed my first proposed milestone - to merge IHT with SnpArrays (see merged pull request here), followed by a series of constructive criticism from my mentor. According to my preliminary benchmark on a dataset of \(2200\) people with \(10000\) SNPs, IHT using SnpArrays as backend runs in about 2.5 seconds and uses around 1.22 GB of memory. Compared to the original implementation which runs in \(\sim\) 2 seconds and uses 23 MB of memory, there is definitely room for memory improvement. In my opinion, this is acceptable as a preliminary implementation to get things working, but we definitely need dramatic memory and speed improvement to scale up to datasets such as the UK biobank. Hopefully my (new) proposed second goal will prove meaningful with respect to this obsticle.

Until next time,