Colm Ryan

And more isogenic CRISPR screens coming out of Toronto @sickkidsto.bsky.social, this time by Mike Tyers and team - congrats! www.biorxiv.org/content/10.6...

starting to suspect that we won't hear the outcome of the Research Ireland Investigators Stage 1 call by the end of January

Thanks to @researchireland.ie for funding, and to the DepMap teams @broadinstitute.org and @sangerinstitute.bsky.social for generating the data that our approach depends on 9/9

A predicted cancer dependency map for paralog pairs Contains predictions for paralog pairs from Kebabci et al 2026.full_SLprediction_matrix.csv.zipDescription: Contains prediction scores generated by the "Full" model.Data Structure* Rows (Index): Cell lines, identified by their DepMap ModelID.* Columns: Paralog gene pairs, formatted as `GeneA1_GeneA2` (e.g., `ARID1A_ARID1B`).* Values: Prediction scores ranging from 0 to 1. Higher values indicate a higher probability of the predicted SL interaction.Usage Notes* File is compressed (.zip) to reduce size. Please unzip to access the .csv matrix.* The first column in the CSV files serves as the index (DepMap ModelIDs).

More broadly, we think this is a useful a foundation for cell line–specific prediction of synthetic lethality beyond paralog gene pairs. Paper's on @biorxivpreprint.bsky.social, predictions available on @figshare.com
figshare.com/articles/dat... 8/9

Using this approach, we generate a genome-scale map of predicted paralog pair dependencies across >1000 cell lines. This resource can be used to prioritise paralog pairs that will cause strong fitness defects in specific contexts (e.g. in HER2 amplified breast cancer models). 7/9

We identify multiple predictive features, including the expression and essentiality of the paralogs themselves and their interaction partners, allowing paralog buffering to be modeled in a cell line–specific network context. 6/9

However, this screening approach is not easily scalable. Instead, here we develop a machine-learning framework to predict pairwise paralog dependencies from existing single-gene CRISPR screening data — both which paralog pairs cause defects and in which specific cell lines. 5/9

One solution is to perform combinatorial CRISPR screens focused on paralogs, as has been done for subsets of paralog pairs in relatively small numbers of cell lines (<30). This has been very informative and has revealed context-specific paralog-pair dependencies. 4/9

A limitation of the DepMap is that it is currently based on single-gene perturbation screens. This is a problem for the ~70% of human genes with paralogs (duplicates). Because pairs of paralogs often share functions, many only cause a fitness defect when perturbed in combination. 3/9

The DepMap enables the discovery of genetic vulnerabilities associated with specific biomarkers and can inform drug development (e.g. identifying WRN as a vulnerability in microsatellite instability–high cells has led to multiple clinical trials of WRN inhibitors). 2/9

New paper from Narod Kebabci – “A predicted cancer dependency map for paralog pairs” www.biorxiv.org/content/10.6...

Background: The Cancer Dependency Map from @depmap.org is a fantastic resource that characterises genetic dependencies at genome-wide scale across ~1,000 cancer cell lines. 1/9

www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.6...

all slightly different gene sets, but a lot of DDR factors

Uncovering genetic interactions in the DNA repair network in response to endogenous damage and ionizing radiation Genomic integrity relies on a complex network of DNA damage response (DDR) pathways that repair endogenous and exogenous lesions, yet how individual f…

Another genetic interaction map of DNA repair factors! I think that makes 5 GI maps (including preprints) of DNA Damage / Repair related factors? Potentially enough to do some systematic comparisons. www.sciencedirect.com/science/arti...

thanks!

Great new work from @colmr.bsky.social predicts a cancer dependency map for paralogs: doi.org/10.64898/202....

New preprint on technologies to scale up CRISPR screens.

We use them to map 665,856 pairwise genetic perturbations and outline a path to comprehensive interaction mapping in human cells.

We also introduce an approach for cloning lentiviral libraries with billions of elements.

In remembrance of Peer Bork  | EMBL EMBL and its community are deeply saddened by the death of Peer Bork, the organisation’s Interim Director General.

very sad news. Peer Bork was one of the leaders of our field, a wonderful scientist, and he's much too young to be gone. www.embl.org/news/embl-an...

Photo containing the following text: "Their paper was published a year before Guyon defended her Ph.D. thesis, for which she had tested numerous algorithms for linear classification—but none of these was an optimal margin classifier, meaning the algorithms found some linear boundary, not necessarily the best one. Guyon could have used Krauth and Mézard's algorithm to implement an optimal margin classifier; she didn't. "One of the examiners of my Ph.D. asked me why I did not implement the algorithm of Mézard and Krauth and benchmark it against the other things I was trying. I said, 'Well, I didn't think it would make that much of a difference,'" Guyon told me. "But the reality is that I just wanted to graduate, and I didn't have time."

Enjoying 'Why Machines Learn' by Anil Ananthaswamy, including this anecdote highlighting that there's always more that *could* be in a PhD but that you have to draw a line somewhere. Guyon here is Isabelle Guyon, who was later key to the development of SVMs (especially the kernel trick)

yay! top of the to-read pile for January!

Functional modules predict cancer-relevant genetic interactions in mammalian cells Genetic interactions can reveal gene function and identify cancer-relevant synthetic lethals, but systematic mapping in human cells is constrained by inefficient reagents, vast combinatorial search sp...

Merry Christmas, genetic interaction nerds:

www.biorxiv.org/content/10.6...

Reduced PRC2 function causes asparaginase resistance in T-ALL by decreasing WNT pathway activity Key Points. Reduced PRC2 function in T-ALL is associated with asparaginase resistance that is linked to reduced WNT/STOP pathway activity.Asparaginase resi

Great to see this out, collaboration with the Bond lab @sysbioire.bsky.social to understand consequences of EZH2 mutation in T-acute lymphoblastic leukaemia (T-ALL). Congrats to @lefeivret.bsky.social @cosmintudose.bsky.social and other authors not on bluesky!

ashpublications.org/bloodadvance...

Replacing my still functioning 2017 Mac because it's no longer compatible with our two factor authentication software (preventing me logging on to any work related system). This really doesn't seem optimal.

Postdoctoral fellow - Petsalaki Group - NetworkCommons Your group The Petsalaki, Saez-Rodriguez (EMBL-EBI) and Korcsmáros (Imperial College) groups develop cutting-edge computational and AI approaches to unravel cellular signalling and gene regulation fro...

#Postdoc alert! We are looking for motivated individual with experience in computational systems biology to join our NetworkCommons initiative to benchmark and democratise network contextualisation methods across multiple applications. Deadline Dec 17th link here: tinyurl.com/yrrzxa94

Last chance to register. Hurry Up!

Anyone with a lower h-index than me is ineffectual, while anyone with a higher h-index is just better at playing the game and cutting corners. I'm sorry, that's just how numbers work!

AEBP2-Directed H3K27me2 Defines a Specific Vulnerability in EZH2-mutant Lymphoma The catalytic subunit of Polycomb Repressive Complex 2 (PRC2), EZH2, is recurrently mutated in 25% of diffuse large B-cell lymphomas (DLBCL), causing increased H3K27me3 and decreased H3K27me2 levels. ...

🧵1/Exciting news in cancer epigenetics! Our latest research, "AEBP2-Directed H3K27me2 Defines a Specific Vulnerability in EZH2-mutant Lymphoma", is now available on www.biorxiv.org/content/10.1.... Here's a thread summarizing our findings!👇
#CancerResearch #Epigenetics #Chromatin #Lymphoma

I've come to the realisation that if I don't have time to review a paper when the request comes in, I probably won't have time to review it in 2 weeks either. The imaginary less busy future never seems to arrive within 2 weeks. Therefore I should accept and review within a few days or decline.

We still can’t predict much of anything in biology Biology is hard. Yes, even for AI.

Biology is much more complicated than most non-biologists can imagine. And AI is not going to change this anytime soon.
blog.genesmindsmachines.com/p/we-still-c...

We think this work will be useful for 1) those choosing an existing method to score a genetic interaction screen and 2) those developing new genetic interaction scoring systems as it provides a framework for benchmarking. Thanks to @researchireland.ie for funding!

We find that no single method performs best across all screens, but one (Gemini) performs well across most and has an available R package (Gemini).