Lightnews — Scholar-powered news

Mike-ghoul Half-man

@michaelhoffman.bsky.social

16K followers 1.6K following 4.5K posts

Chair, Computational BIology and Medicine Program, Princess Margaret Cancer Centre, University Health Network. Associate Professor, Medical Biophysics, University of Toronto. Disclosures: https://github.com/michaelmhoffman/disclosure/

github.com

Posts Media Videos Starter Packs

Pinned

Mike-ghoul Half-man @michaelhoffman.bsky.social · Nov 12

I'm at Princess Margaret Cancer Centre and University of Toronto. We have three major interests:

1/Computational methods and #ML for gene regulation data 🧬🖥️
2/Liquid biopsy beyond simple genetic variants (epigenomic, immunogenomic)
3/Reproducibility and robustness of biomedical analysis (esp. #ML)

2 6 57

Mike-ghoul Half-man @michaelhoffman.bsky.social · 54m

Now: Xiangyu Ye: Single-cell, trimodal omics atlas of ileal and colonic mucosal T cells in Crohn's disease highlights CD4⁺ regulatory T cell plasticity in inflammation progression #ASHG25 🧪🧬🖥️

1 1

Mike-ghoul Half-man @michaelhoffman.bsky.social · 1h

RL: Integration of genetic associations and Perturb-seq limitations: two-stage approach ignores trait of interest while extracting programs from Perturb-seq. For that reason, didn't identify trait-relevant progras when applied to Jurkat T cells and lymphocyte count. #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 1h

Now: Romain Lopez: PhenoBridge: a unified framework for interpreting molecular effects on complex phenotypes through integration of Perturb-seq and phenotypic data #ASHG25 🧪🧬🖥️

1 1

Reposted by Mike-ghoul Half-man

Marit Stiles @maritstilesndp.bsky.social · 1h

800,000 people are out of work — and what’s their Premier doing? Eating ice cream, pouring out liquor for the cameras, and playing Batman, bragging that he wanted to beat someone up at Home Depot.

What will it take for him to stop chasing the stunts and the theatre? A million people unemployed?

7 10 41

Reposted by Mike-ghoul Half-man

Luke Plunkett @lukeplunkett.com · 8h

kids teacher just sent this home as part of an assignment he's doing and i want to smash every computer at the school

120 1.3K 5.4K

Mike-ghoul Half-man @michaelhoffman.bsky.social · 1h

JM: Showing results using systematic targeting and inhibition of noncoding GWAS loci with single-cell sequencing (STING-seq) #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 1h

Now: @johnomix.bsky.social (JM): Systematic targeting and inhibition of bone mineral density GWAS loci in single cells identifies novel regulatory networks for human bone formation #ASHG25 🧪🧬🖥️

1 1 2

Mike-ghoul Half-man @michaelhoffman.bsky.social · 1h

In the session on Seq-ing function with single cells and CRISPR perturbations. Moderators Ethan Fenton and Siming Zhao #ASHG25 🧪🧬🖥️ https://go.pheedloop.com/events/EVEAUQQRGTVJR/sessions/SES05JL3KQ6OCBM45

Mike-ghoul Half-man @michaelhoffman.bsky.social · 2h

I don't know about the trigger point stuff but knots are literally palpable

Reposted by Mike-ghoul Half-man

Shawn Micallef @shawnmicallef.bsky.social · 3h

"Safe Parkside" filed a Freedom of Information request to get top speeds from the ill-fated & much abused Parkside camera. The speed limit is 40km/h.

Killers, all of them, with the ear & heart of Premier Doug Ford.

9 57 120

Mike-ghoul Half-man @michaelhoffman.bsky.social · 15h

"There is no such thing as sauciety." —Margaret Thatcher, probably #ASHG25

2 2

Mike-ghoul Half-man @michaelhoffman.bsky.social · 16h

The Dog Murderer-in-Chief.

Mike-ghoul Half-man @michaelhoffman.bsky.social · 16h

Suggestions for a book about Boston that I can send or bring back to a 5-year-old? #ASHG25

1 1

Mike-ghoul Half-man @michaelhoffman.bsky.social · 17h

IB: Phase 2: test them as referral support tools, with and without supplemental education, to assess impact on referral appropriateness. #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 17h

IB: Only 14 clinical geneticists completed the survey. The 40 vignettes did not capture the full complexity and variability of real-world genetic cases. Scores might have had bias. LLM models not fixed so reproducibility hard #ASHG25

1 1

Mike-ghoul Half-man @michaelhoffman.bsky.social · 17h

IB: LLMs similar. #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 17h

IB: Overall diagnostic performance between LLMs and geneticists: LLMs 66% correct, geneticists 53%; p=0.03 #ASHG25

1 2

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: Score the report.
2: correct diagnosis ranked first
1.5: top answer incorrect, correct diagnosis listed first in differential
1: top answer incorrect, but correct diagnosis listed second or third in differential.
0.5: correct diagnosis listed later.
0: correct diagnosis not in result. #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: Study design: 40 real-world genetic case vignettes. Recruited 300 board-certified geneticists. Compared with ChatGPT 3.5, LLaMa, and Gemini [result: accrual of board-certified geneticists easier than independent LLMs] #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: CResearch question: canLLMs provide comparable results to clinical geneticists? #ASHG25

1 3

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: LLMs don't understand medicine or context the way we do. They always need human supervision. #ASHG25

1 2

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: A large language model is a type of computer program that has read huge amounts of text and learned to recognize patterns in language #ASHG25

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

IB: Shortage of geneticists → long wait times for patients. Many physicians from other specialties (e.g. nephrology, cardiology) use genetic results but are under-supported and underprepared → delay in getting results #ASHG25

2 1

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

Now: Ivy Bethea: Comparing the diagnostic capability of large language models and clinical geneticists #ASHG25 🧪🧬🖥️

PheedLoop

meetings.ashg.org

1 2

Mike-ghoul Half-man @michaelhoffman.bsky.social · 18h

Q: Colocalization analysis between genders [actually sexes]? EW: We didn't colocalize but we did do LD comparison of lead variants. #ASHG25

1 1