Tessera

Liuyang Wang

@wallacewly.bsky.social

Human genetics; Statistical genetics; GWAS; Infectious disease; @DukeMGM; Father; Husband @DukeMedSchool https://scholars.duke.edu/person/liuyang.wang

2 Posts42 Follower 94 Folgt

Liuyang Wang @wallacewly.bsky.social · last mo.

What happens when AI meets human genetics? We integrate single-cell sQTL mapping with fine-mapping and deep-learning splicing prediction to uncover disease causal genetic variants and their mechanisms. Read our new study in Genome Biology: link.springer.com/article/10.1...

Integrating single-cell sQTL mapping with deep-learning splicing prediction identifies causal variants under influenza infection - Genome Biology

Background Realizing the full potential of human genetics requires identifying causal variants and genes underlying association signals. Molecular quantitative trait locus (molQTL) analyses, such as e...

link.springer.com

David Tobin @tobinlab.bsky.social · 2mo ago

New facets of eicosanoid signaling in TB: crosstalk with a specialized population of granuloma-associated fibroblasts. Happy to share new work led by Erika Hughes, with @charliejpyle.bsky.social and many other wonderful collaborators not on BlueSky.

Host eicosanoid signals define a granuloma fibroblast population that coordinates mycobacterial containment

Hughes et al. find that the TB susceptibility gene lta4h contributes to the recruitment and differentiation of a discrete set of fibroblasts that populate the periphery of mycobacterial granulomas in ...

cell.com

Liuyang Wang @wallacewly.bsky.social · last yr.

Paired spatial snRNA-seq and scRNA-seq from 4 TB patients are alive at Broad Single Cell Portal, you can look up your genes in seconds :)

Charlie Pyle@charliejpyle.bsky.social · last yr.

You can now explore the mRNA expression of your favorite TB gene of interest in our recently published human pulmonary M.tb granuloma single-cell dataset at the Broad Single Cell PORTAL! @broadinstitute.org @wallacewly.bsky.social singlecell.broadinstitute.org/single_cell/...

Nature Reviews Genetics @natrevgenet.nature.com · last yr.

Evolution of splicing model architectures go.nature.com/4eweliE Figure from our recent Review: From computational models of the splicing code to regulatory mechanisms and therapeutic implications (free to read here: rdcu.be/dVNV4)

a, Early models built sequence motifs to describe the consensus sequences of individual core splicing elements, such as splice sites (SSs) and intronic and/or exonic enhancers and silencers. Statistical and machine-learning models were built to output the probability of a novel sequence acting as a core splicing element. The sequence logos shown for 5′SS and 3′SS were generated from Human hg38 RefSeq annotations (code available at https://www.github.com/ulelab/splicelogos). b, As our understanding of splicing mechanisms progressed, expert-selected features were extracted from sequences and used to train integrative models to predict splicing outcomes. c, With the advent of deep-learning, models could jointly learn features directly from raw sequence input. Although theoretically, sequence context could be as large as shown in part d, in practice smaller windows of up to 30 kb have been used. d, Supervised models with convolutional and transformer layers produce multimodal genome-wide data. These models use a much larger sequence context and can predict genome-wide data including RNA sequencing coverage, which can be further processed to evaluate splicing. e, By learning how to reconstruct partially masked genomic sequences across multiple species, self-supervised masked language models capture evolutionarily conserved sequence elements and their functional context in a very generic and flexible fashion. The informative numerical representations obtained by large language models can be used for splicing prediction tasks. Here 3′SS within different sequence contexts from multiple species are shown aligned for easier interpretation, but in practice sequences do not have to be aligned. Current masked language models with application to splicing use variable context windows from 1,000 to 1 million base pairs; however, it is currently unclear whether larger context windows confer better performance

Charlie Pyle @charliejpyle.bsky.social · last yr.

Proud to present our latest work! @tobinlab.bsky.social @wallacewly.bsky.social @gregorylab.bskyverified.social Human TB granuloma transcriptional blueprints show they are dominated by SPP1 macrophages. spp1 is essential for granuloma formation in zebrafish. journals.asm.org/doi/10.1128/...

Paired single-cell and spatial transcriptional profiling reveals a central osteopontin macrophage response mediating tuberculous granuloma formation | mBio

Tuberculosis is the world’s most deadly single-pathogen infection. Its causative bacterium, Mycobacterium tuberculosis, sickens over 10 million people annually. Mycobacterial granulomas are the pathol...

journals.asm.org