iiinova — where your digital self keeps evolving

Jonathan Sebat @sebatlab.bsky.social · 27/09/2025

As a novel approach to causal inference, we will use collider effects to find genetic evidence to SUPPORT a causal role of environmental exposure. If E has a strong effect that is uncorrelated with G in the population, genes will explain LESS of the variance in the explosed group!

110

Jonathan Sebat @sebatlab.bsky.social · 27/09/2025

Our 2022 paper in NG illustrates one example: the observed association of parental age with ausism consists of a mixture of causal pathway E → de novo mutation → ASD; and noncausal pathway Parental age ← polygenic score → ASD

100

Jonathan Sebat @sebatlab.bsky.social · 27/09/2025

We will use a multi-pronged approach to causal inference using data on G and E in >2 million people, including association, MR and GxE exploring the interplay of exposures with rare and common variation that contributes to ASD, and we will rigorously evaluate rGE in family and population

100

Jonathan Sebat @sebatlab.bsky.social · 27/09/2025

THE major challenge in epidemiological studies of the environment in autism is the pervasive passive correlation of genes and environment (rGE) that happens in the population, also known horizontal pleiotropy. There are causal effects to be found, but we need to separate correlation from causation

110

Jonathan Sebat @sebatlab.bsky.social · 16/08/2025

Bad ass mom of the year 🤘. This was NOT an easy coffee ride. This was the Saturday Cadence Cyclery ride in SD.

020

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

Power is limited in N=243. But the signal is quantifiable. Rare SVs, TRs and rare SNVs combined explained 7.6% of the heritability. We will need larger samples to get precise estimates, but the combined common and rare is starting to reach a substantial fraction.

100

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

Gray zone carriers typically look like panel E. Only FMR1 methylation is skewed. X chromosome inactivation is not skewed. The one exception (panel D) was a girl that carried an X-linked dominant mutation in DDX3X a gene that escapes XCI and, for unknown reasons, is know to cause skewed XCI

130

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

Skewed methylation was a characteristic of all "gray zone" alleles of FMR1 (>35 repeats). But in all but one sample, the skewing was specific to FMR1. X chromosome inactivation was NOT skewed. 10/N

100

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

In the fragile X repeat FMR1, we went full long-read nerd with PHASED repeats AND methylation. Phased reads in one female (REACH365) shows random methylation (red) on the X. In another female (REACH561) with an expanded 49 repeat allele, methylation was totally skewed to expanded haplotype H2 9/N

110

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

Where long reads really shine is when you need to map out and assemble large complex SVs. In doing so, we found a class of duplication/deletion events (TAN-DUP-DEL and INV-DUP-DEL) events. They also produce distinct signatures in short read coverage data

130

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

De novo SVs have been bread-and-butter of ASD genetics, and LRs detect novel de novo and somatic MOSAIC coding SVs. In this example, a de novo in-frame duplication, was present on most (but not all) reads on the H2 haplotype in the offspring and coverage also showed copy number of 2.5 instead of 3.

120

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

Panel C (coding SVs in constrained genes) are a class variants with well-established associations with ASD. To maximize detection of these, one needs a combination of short reads and long reads. For large (>50 bp) tandem repeat variants, long reads alone do a good job 4/N.

100

Jonathan Sebat @sebatlab.bsky.social · 24/07/2025

a susbstantial fraction of SVs are only detetable with LRs. However, there is also a significant fraction of SVs that are detectable only with Illumina WGS, particularly large CNVs that are detectable based on coverage. (A) All SVs (B) coding SVs (C) coding SVs in constrained genes (pLI>0.9).

130

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

While not attributable to the major loci, patterns are consistent within them. When you look closely at cell-type specific gene expression in DUP 16p11.2 or DEL 22q11.2, you begin to appreciate how the effects of a CNV can be mediated through distinct pathways across a variety of cell types

100

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

Factor F2 was driven by differential DUP effects in neuronal vs non-neuronal cell types in mood disorders. F3 by differential spatial patterns of DEL effects in MDD and ADHD. The 3 factors, factor loadings and factor scores remain concordant after we remove all 18 genome-wide significant loci

100

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

As a reminder, DEL effects are concentrated in visual, auditory, sensorimotor. DUP effects are more frontal. Dose dependent effects in SCZ appear to show differential patterns at multiple levels synaptic/regulatory, excitatory/inhibitory, sensory/association

110

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

The major factor F1 represents dose-dependent effects in SCZ characterized by DUP effects in Reg pathways in postnatal Exc. neurons, neurovascular, microglia - and divergent DEL effects in synaptic, Ca+, axon guidance in postnatal Inh. neurons as well as early fetal neurons.

100

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

To elucidate where gene-dosage effects converge, we performed exploratory factor analysis of functional gene sets to identify latent factors that correspond to different gene-trait relationships. In all Dx's except ASD DEL and DUP effects form distinct components in distinct neural processes

110

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

To characterize how gene dosage effects are distributed in the brain, we performed GSBA on the the intersections of pathways, cell types, and brain regions. We then evaluated which levels of biological organization best explain variation in gene-set effects in mixed effects models 11/N

110

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

Team members @kkumar.bsky.social and Sebastien Jacquemont devised some interesting ways to apply GSBA to gene sets highly expressed in 180 cortical brain regions. When gene set associations are projected on the brain map, Psych Dxs differ in the spatial distribution of DEL and DUP effects

110

Jonathan Sebat @sebatlab.bsky.social · 17/07/2025

Considering 1 level of biological organization at a time, the pathways and cell types associated with ASD, SCZ and other disorders are exactly as we expected. But we begin to see DIVERGENCE between Dx's when we stratify by DOSAGE. In SCZ, DUP→core regulatory pathways, DEL→synapse, calcium...

110

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

Given that CNVs exhibit dose dependent effects on psychiatric traits, we developed a framework to characterize how gene-Dx relationships are mediated by pathway, cell type, brain regions. We begin with a statistical genetic approach Gene Set Burden Analysis (GSBA)

110

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

And gene findings in schizophrenia were also, you guessed it, synaptic, chromatin, cell signaling etc. 3/N

110

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

From the early days of autism genetics PMID: 24768552, we marveled at all the "functional convergence". The genes that were identified by CNV, whole exome studies were clearly pointing at pathways that regulate early brain development! including synaptic, chromatin, cell signaling 2/N

100

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

While finding new genes is not hard, it's still cool! We find a genome-wide significant association of SMYD3 duplications with SCZ. SMYD3 has been a target of interest in oncology for years. A variety of SMYD3 inhibitors have been developed. Possible therapeutic implications for psychiatry.

120

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

Likewise DEL and DUP of a locus is associated with contrasting spectrum of Dx's. Del 22q11.2 associated with SCZ and ASD. DUP is *protective* for SCZ and it increases depression. At the 16p11.2 proximal DUP is associated with adult onset psychiatric Dx. DEL is NOT.

110

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

The diverging associations of CNVs with different Dx is a reflection of dose-dependent relationships of genes with psychiatric traits. When you plot the effect size for deletion and duplication of the same genes, they show clear linear dose response for some Dx's such as SCZ. But not in ASD

110

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

As expected there's alot of overlap. Some alleles (DUP 22q11 A-D, Dup 22q11 B-D, DEL 15q11.2-CYFIP1) are positively associated with "risk" of one Dx and negatively associated with another. Yes bro, rare variants are protective. BUT "Protection" from one Dx is "Susceptibility" to another

122

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

A common thread between both papers is that pleiotropy is not simply an interesting or confounding phenomenon. PLEIOTROPY IS THE POINT. Many different psychiatric traits share common genes. But we didn't just link genes to trait A, B, C, D. We quantified differential effects of gene dosage on each

173

Jonathan Sebat @sebatlab.bsky.social · 16/07/2025

The meta-analysis of CNV calls across a variety of samples and genotyping platforms is a non-trivial but solvable problem. Main consideration that meta-analysis should stratify by genotyping platform. Across 6 Dxs, there were 35 genome wide associations at 18 loci. Pleiotropy was ENTIRELY expected

110

Jonathan Sebat @sebatlab.bsky.social · 09/04/2025

IF U SEE THIS POST A BEACH PHOTOGRAPH FROM UR TELEPHONE

210

Jonathan Sebat @sebatlab.bsky.social · 08/02/2025

If you are wondering “how is this possible?”. Officially it’s not, so this is obviously an attempt to squeeze Universities and get some quick leverage over them

020

Jonathan Sebat @sebatlab.bsky.social · 11/01/2025

Who does Denmark think they are fooling? These all taste the same.

5133

Jonathan Sebat @sebatlab.bsky.social · 05/01/2025

I Viktor, nice to see that this study addressed genetic confounds. When I look at those results I don't see many effects that persist in the sibling control analysis... at least until the results are compared to a single antionvulsant (Lamotrigne). Is it standard to use this group as the baseline?

100

Jonathan Sebat @sebatlab.bsky.social · 30/12/2024

I stand corrected, the sample is matched to a control group of students who start a masters at the same time. One of the biggest factors predicting medication uses (relative to matched controls) was sex. There is no effect in males relative to controls. This is a red flag

100

Jonathan Sebat @sebatlab.bsky.social · 29/12/2024

Why not correct for patterns of med use for non-PhDs? www.jabfm.org/content/jabf...

120

Jonathan Sebat @sebatlab.bsky.social · 01/12/2024

After the bomb cyclone passed, Thanksgiving in Big Sur was amazing

030

Jonathan Sebat @sebatlab.bsky.social · 13/11/2024

Sorry I missed the zoom call this morning

020

Jonathan Sebat @sebatlab.bsky.social · 20/09/2023

Very proud of our PhD candidate Omar Shanta. He has been a driving force of the @PGCgenetics CNV analysis group. Mark your calendars and DM me for the zoom

010