Evan Maxwell
@evankmaxwell
ID: 786365399470182401
13-10-2016 00:38:16
78 Tweet
44 Followers
318 Following
Jeffrey G. Reid Variant calling essentially "solved", but edge cases are complicated. Retraining already well-performing models leads to further improvement. Next steps: More difficult training curriculum #ASHG19
Jeffrey G. Reid #ASHG19 Used Regeneron's refined exome seq. Re-trained DeepVariant to make own model. Warmstart from WES model, converges quickly to new tuned model. Large improvement to indels, difficult regions (exome buffer). Reduce Mendelian violation, esp outside confident region.
Pls RT! Exciting 2020 internships (one with me) in ML and Genetics for health in the method dev team at Regeneron Genetics Center - team led by Jonathan Marchini (with @codushlaine, etc). Work with me on utilizing pathway information to discover drug targets.👇 careers.regeneron.com/job/REGEA00261…
Here is a new paper from the Regeneron Genetics Center biorxiv.org/content/10.110… A new, fast ML method for whole genome regression. The REGENIE website is here rgcgithub.github.io/regenie/ with free source code. Amazing work from @jojoladisco Leland Barnard Evan Maxwell 1/10
Huge extra-special thanks to the brilliant @Databricks team who have really made this possible -- most especially Henry Davidge Henry Davidge Karen Feng Karen Feng Kiavash Kianfar @KiavashKianfar and led by Frank Austin Nothaft Frank Austin Nothaft -- a good friend and great collaborator!
We also have to thank everyone from the UK Biobank Exome Sequencing Consortium (UKB-ESC) AbbVie Alnylam Pharmaceuticals AstraZeneca Biogen Science At BMS Pfizer Inc. Takeda -- who partnered with us Regeneron to generate the exome data, do this analysis, and make these results public.