Maria Chikina

Voters expressed genuine grievances, and while change is rarely immediate, this outcome may still signal shifts that align the economy more closely with the needs of the people. And that’s an aspiration we can all share.

Democracy, though not always a direct reflection of our ideals, remains a system responsive to collective concerns, even if indirectly.

Many are feeling the weight of the election's outcome. As we process this, let's focus on what we can control: our work, connections with friends and family, and sources of personal fulfillment.

Our work benchmarking proportion estimation methods (genomebiology.biomedcentral.com/articles/10.11…) finds BayesPrism (nature.com/articles/s4301…) among the top performers. We introduce a fast re-implementation InstaPrism: same accuracy, orders of magnitude faster! academic.oup.com/bioinformatics…

A 🧵 on some of my intuitions/priors about the genetics of complex and molecular traits in humans (i.e. what I think of as typical), largely motivated by GWAS/QTL studies over the past decade [citing papers with nice figures where possible]

Can someone explain Fig 2 of nature.com/articles/s4157…. Why do some nascent RNA techniques give signal past the transcription end site and others don't.

@anshulkundaje But the slow and diligent approach is not incentivized. How might we be able to solve that problem?

The goal of the research and science is to actually try to solve problems, not to bang out papers fast, first and in the most shiny journal (despite what u may have been told & trained to do). Seriously, we must all try to do better. 10/10

Ok I'm going to say it out loud. I think it is a disservice to any field to publish papers (and especially high-profile ones) hastily to be "first in line" that have not gone through sufficient basic due diligence by the authors to back up their primary claims. 1/

@anshulkundaje @sara_mostafavi @LXandR_ We are not saying you need to replace the reference genome with another one but that you need more genome/output pairs. Clearly in the limit of variant memorization this *will* increase prediction accuracy but perhaps it can also be used to get a more biochemically aware model.

@sara_mostafavi @ChikinaLab @LXandR_ Much needed analysis. Thank you! One thing I will slightly disagree with (for now), is the claim that reference genome training is the main reason for the failures. 1/

Interested in using deep learning to predict the impact of genetic variation on gene expression? In our pre-print we investigate how well this can be done currently . Joint work with wonderful colleagues @ChikinaLab, Alex (@LXandR_ ), and others. biorxiv.org/content/10.110…

Hopefully this in depth evaluation will pave the way for more accurate personalized gene expression models. We will surely need more data and likely different models and training paradigms.

In our new paper embopress.org/doi/full/10.15… we show that unlike gene expression methylation changes persist for months after infection and could be used to predict infection timing. Great effort by @WayneMao1.

If you need to deconvolve bulk RNAseq data and like BayesPrism (tinyurl.com/bdzddefv) consider InstaPrism (tinyurl.com/2p8s7rcj). Same results but much faster.

We just posted a new preprint on benchmarking deconvolution methods. With an appropriately realistic simulation we get highly counter-intuitive results. The problem is far from being completely solved! biorxiv.org/content/10.110…

Here Today, Gone Tomorrow: How (Some) Mammals Lost Their Body Hair new release on lab publication from @KowalczykAmanda and @ChikinaLab healthcare.utah.edu/publicaffairs/… via @UofUHealth

Our new paper with @NathanClark111 and @KowalczykAmanda looking at mammalian sparse hair phenotype is now online. elifesciences.org/articles/76911

New preprint with @ChikinaLab: "The paradoxical nature of easily improvable evidence" When trying to reason about available evidence which is filtered through the lens of publication bias, what you *don't* see can be as informative as what you do see. 1/2 arxiv.org/abs/2205.14016

When considering doing a correlation analysis, a useful thought experiment is to imagine the conclusion for every possible outcome (+, -, NS). This avoids the trap of only accepting the results that reinforce one’s prior and discarding others with some post-hoc explanation.

This thread on a recent paper from @ProfEmilyOster and others is making the rounds. It is an extremely critical thread employing strong language, questioning the basic judgement and statistical understanding of the authors. A quick, polite thread with some comments on this.1/12

At a time when many comments from just a few months ago have aged badly this timeless quote from Bertrand Russell is both inspirational and comforting. Stick to the facts, respect those that disagree and you are sure to have few regrets.

Happy to announce our recent paper with @ChikinaLab, @jcmattingly, and Alan Frieze on new Markov chain significance tests in the context of redistricting problems. It includes examples from the 2019 Common Cause v. Lewis case in North Carolina. tandfonline.com/doi/full/10.10…

Alpha-ketoglutarate (AKG) is a chemical used by the cell to make amino acids, energy & reverse the "DNA methylation" clock. Feeding 18-month old female mice AKG reduced hair greying, frailty & extended median lifespan 10-16% 💪 Worked less well on boys 😒 tinyurl.com/y6qpyrtu