May 23rd, 2019
The shortest bi-weekly genomics research report in the world. Curated by the Emedgene Research Division.
The AI Corner
Unsupervised learning in medicine! We don’t see this type of ML research often. Instead of training their algorithms with a large data set, these researchers allowed the algorithm to cluster results on its own, looking for hidden groupings. The outcome was 4 distinct clinical phenotypes for sepsis. However, clinical utility is still unknown.
Google AI’s lung cancer screening algorithm published results. The AI performed as well as, and sometimes better than the 6 radiologists it was tested against. This demonstrates the possibility of optimizing and scaling the screening process.
Lastly in the AI corner, is this review attacking explainable AI models that are retrofitted to explain black box machine learning algorithms. Our main take away here was that we should do our best to choose the correct, interpretable models that will allow us to later explain our recommendation.
Nature Machine Intelligence
A Waste of 1,000 Research Papers
That’s not our headline, but the Atlantic’s, dramatically summarizing new findings, or rather un-findings. It was science paper of the year in 2002, but a large study of major depression candidate genes and phenotypes finds little evidence to support previous MDD candidate gene hypotheses.
American Journal of Psychiatry
What can we gain from going long (vs wide)? Study collected longitudinal data on a cohort at risk for type 2 diabetes – genome, immunome, transcriptome, proteome, metabolome, microbiome and wearable monitoring.
GIF Break !
The best data visualization we’ve seen in a while!
The Big T2D Study
Largest known analysis of T2D (type-2 diabetes) exomes. Rare variant contribution is lower than expected and exome-wide significance in drug target genes will require significantly larger cohorts.
Polygenic yay or nay?
A well-written perspective on polygenic risk scores and whether we should be using them now. As usual, it comes down to clinical utility.
New England Journal of Medicine
They had tissues!
With over 2000 tissue samples, authors were able to analyze multi-tissue transcriptome to identify genetic mechanisms underlying neuropsychiatric traits.
Comparing genotyping 23 variants, with NGS of CFTR and NGS + CNV for cystic fibrosis carrier screening. Limiting screening does not improve clinical validity or utility, but reduces detection rate.
Genetics in Medicine
- MYBPC1 associated with an expanded neuromuscular phenotype beyond arthrogryposis Human Mutation
- SHOC2 mutation associated with prenatal‐onset hypertrophic cardiomyopathy Human Mutation
- PAX7 & myopathy Genetics in Medicine
(Ada is our AI engine, she likes to add an interesting new factoid she discovered in every issue)
Rare variants in BNC2 are implicated in autosomal-dominant congenital lower urinary-tract obstruction. Confirmed in unrelated humans, mice and zebrafish knockdowns.
Questions? Comments? Corrections? Please email us at firstname.lastname@example.org