feed.xml

<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.9.2">Jekyll</generator><link href="https://jthlab.github.io/feed.xml" rel="self" type="application/atom+xml" /><link href="https://jthlab.github.io/" rel="alternate" type="text/html" /><updated>2024-04-27T00:22:17+00:00</updated><id>https://jthlab.github.io/feed.xml</id><title type="html">Terhorst lab</title><subtitle></subtitle><author><name></name></author><entry><title type="html">Congrats Yifan!</title><link href="https://jthlab.github.io/2022/08/16/Congrats-Yifan!.html" rel="alternate" type="text/html" title="Congrats Yifan!" /><published>2022-08-16T00:00:00+00:00</published><updated>2022-08-16T00:00:00+00:00</updated><id>https://jthlab.github.io/2022/08/16/Congrats-Yifan!</id><content type="html" xml:base="https://jthlab.github.io/2022/08/16/Congrats-Yifan!.html">&lt;p&gt;Congratulations to Yifan for successfully defending! Yifan did interesting work on several challenging problems related to coalescent hidden Markov models. It’s hard to believe that we’ve been working together for five years! Yifan’s future plans are up in the air but I’m sure he will do well wherever he lands. Good luck Yifan!&lt;/p&gt;</content><author><name></name></author><summary type="html">Congratulations to Yifan for successfully defending! Yifan did interesting work on several challenging problems related to coalescent hidden Markov models. It’s hard to believe that we’ve been working together for five years! Yifan’s future plans are up in the air but I’m sure he will do well wherever he lands. Good luck Yifan!</summary></entry><entry><title type="html">New Paper On Li Stephens</title><link href="https://jthlab.github.io/2022/08/06/New-Paper-on-Li-Stephens.html" rel="alternate" type="text/html" title="New Paper On Li Stephens" /><published>2022-08-06T00:00:00+00:00</published><updated>2022-08-06T00:00:00+00:00</updated><id>https://jthlab.github.io/2022/08/06/New-Paper-on-Li-Stephens</id><content type="html" xml:base="https://jthlab.github.io/2022/08/06/New-Paper-on-Li-Stephens.html">&lt;p&gt;Today we posted a &lt;a href=&quot;https://www.biorxiv.org/content/10.1101/2022.08.03.502674v1&quot;&gt;new paper&lt;/a&gt; on the Li-Stephens (LS) haplotype copying model. LS is a beautiful technique for approximating the likelihood of a sample of whole genome sequence data, which is otherwise very hard to evaluate because of the astronomically large number of ancestry scenarios that could have generated the data. 
If you work in stat gen, chances are good you have used this model, perhaps without realizing it, since it underpins a lot of other extremely popular computational methods, including IMPUTE, SHAPEIT, EAGLE, fastPHASE, and maCH. For a brief summary of the method and all its applications, see this &lt;a href=&quot;https://academic.oup.com/genetics/article/203/3/1005/6066789&quot;&gt;review paper&lt;/a&gt; by Yun Song, or just read the &lt;a href=&quot;https://academic.oup.com/genetics/article/165/4/2213/6050566&quot;&gt;original paper&lt;/a&gt;, which is very nice and not too technical.&lt;/p&gt;

&lt;p&gt;As with any pop gen method, the output of the LS algorithm depends on some population genetic parameters, in this case the rates of recombination and mutation. We were curious how the output of the LS algorithm changes in response to perturbations in these parameters. Answering this question rigorously turns out to be statistically and mathematically interesting, with connections to convex analysis and changepoint detection. It’s also practically important since, presumably, the performance of the many algorithms founded upon LS depends on setting these parameters in an optimal way.&lt;/p&gt;

&lt;p&gt;Our contribution here is to derive an efficient algorithm for uncovering &lt;em&gt;all&lt;/em&gt; of the possible solutions to the LS algorithm in a single run. This is akin to what the LARS algorithm does for the LASSO problem (\(\ell_1\) penalized regression). Unlike regression, in practice, the “answer” to LS is not known (and potentially not even defined), but using simulations and our algorithm, we can study how different parameterizations affect the ability to do certain tasks like imputation and phasing.&lt;/p&gt;

&lt;p&gt;One interesting finding from our paper is that using the “population-scaled” rates of recombination and mutation might actually be suboptimal, at least for some problem instances. In the graphic below, we compared the distribution of switch errors committed by a haplotype phasing algorithm over repeated simulations. The orange histogram is the distribution when the population-scaled values are used. The blue histogram is what we get 
by using our algorithm to figure out the best possible settings for these parameters in each run. At least on the simulation considered here, there is a gap of a few percentage points in switch error that could possibly be erased by parameter tuning.&lt;/p&gt;

&lt;p&gt;&lt;img src=&quot;/assets/img/blog/2022-08-06/switch_error.png&quot; alt=&quot;phasing error&quot; /&gt;&lt;/p&gt;</content><author><name></name></author><summary type="html">Today we posted a new paper on the Li-Stephens (LS) haplotype copying model. LS is a beautiful technique for approximating the likelihood of a sample of whole genome sequence data, which is otherwise very hard to evaluate because of the astronomically large number of ancestry scenarios that could have generated the data. If you work in stat gen, chances are good you have used this model, perhaps without realizing it, since it underpins a lot of other extremely popular computational methods, including IMPUTE, SHAPEIT, EAGLE, fastPHASE, and maCH. For a brief summary of the method and all its applications, see this review paper by Yun Song, or just read the original paper, which is very nice and not too technical.</summary></entry><entry><title type="html">Congrats Caleb!!</title><link href="https://jthlab.github.io/2022/04/27/Congrats-Caleb!!.html" rel="alternate" type="text/html" title="Congrats Caleb!!" /><published>2022-04-27T00:00:00+00:00</published><updated>2022-04-27T00:00:00+00:00</updated><id>https://jthlab.github.io/2022/04/27/Congrats-Caleb!!</id><content type="html" xml:base="https://jthlab.github.io/2022/04/27/Congrats-Caleb!!.html">&lt;p&gt;Congratulations to Caleb for successfully defending and graduating! It was great
to get a chance to meet Caleb’s family, and he was my first student to boot! Caleb
is off to New York City to be a data scientist for Lyft. He will be missed!&lt;/p&gt;

&lt;p&gt;&lt;img src=&quot;/assets/img/blog/2022-04-28/IMG_0261.jpg&quot; alt=&quot;caleb and jonathan&quot; /&gt;&lt;/p&gt;</content><author><name></name></author><summary type="html">Congratulations to Caleb for successfully defending and graduating! It was great to get a chance to meet Caleb’s family, and he was my first student to boot! Caleb is off to New York City to be a data scientist for Lyft. He will be missed!</summary></entry></feed>