LIBD rstats club

Talk submission video for Posit Conf 2026 #RStats

«Organizing your data science team to be efficient, adaptive to new developments, and resilient to turnover rate»

URLs:
1. lcolladotor.github.io/bioc_team_ds/
2. research.libd.org/rstatsclub/
3. github.com/LieberInstit...

#OpenSource #Bioconductor

RDM Weekly - Issue 031 A weekly roundup of Research Data Management resources.

Issue 31 of RDM Weekly is out! 📬

➡️ OS resources in Comm Sciences @csdisseminate.bsky.social
➡️ RDA 25th Plenary Meeting Programme @researchdataall.bsky.social
➡️ Ask for R Help by Creating a Small Reproducible Example @libdrstats.bsky.social
and more!

rdmweekly.substack.com/p/rdm-weekly...

@lieberinstitute.bsky.social @jhubiostat.bsky.social

lcolladotor.github.io/rnaseq_LCG-U... is looking ready for next week!

We'll have the most guest presenters ever this year with @meli-mq99.bsky.social @gabrielramirezv.bsky.social @mariagtac.bsky.social & @daiannaglez.bsky.social 🙌🏽

See you soon at LCG-UNAM! 🇲🇽

#RStats @bioconductor.bsky.social

In a recent journal club, I covered the SOFA paper by @wkhuber.bsky.social and colleagues. It's especially cool how you can guide the factorization approach with known covariates.

Check out their preprint (doi.org/10.1101/2024...) and our discussion (youtu.be/WKs-g4lXzZs)!

A colleague Geo @geopertea.bsky.social found out a cool trick to get Positron working with full support at our computing cluster JHPCE (remote file explorer, Positron assistant seeing the remote R session). Game changer for my work at LIBD! Check out his demo video (youtu.be/CnmaJG8VPp8) #positron

noted that DuckDB won't perform worse than dplyr at any point ✅, and it's pretty easy to use 👍, so he plans to keep using duckplyr more often 🚀.

He wrapped up by saying the benchmarks were interesting 📈, and despite the mixed results, he still expects to be using duckplyr a lot going forward! 💪😊

9 hours ⏰😫, and he was trying to optimize it.

Surprisingly, this particular use case didn't benefit much from DuckDB 🦆, leading to his main takeaway: performance is highly dependent on what you're doing 🎯. However, he

@nick-eagles.bsky.social concluded that scratch space location matters significantly for performance 💾⚡. He explained his motivation: he had a real use case with hundreds of millions of rows 📊 that originally took

Full length video: youtu.be/ikumz_QLZiE
#RStats #dplyr #duckplyr #rpolars #data.table

He noted that most benchmarks focus on speed ⚡, but emphasized that memory efficiency 🧠 is another important variable to consider - particularly highlighting that data.table tends to be more memory efficient than dplyr 💾✨.

#RStats #dplyr #duckplyr #rpolars #data.table

@nick-eagles.bsky.social introduced his presentation on benchmarking #dplyr alternatives 📊💻. He planned to review public benchmarks of various data manipulation packages 📦 as well as his own custom benchmark results 🔬

Full length 🎥: youtu.be/ikumz_QLZiE
#RStats #dplyr #duckplyr #rpolars #data.table

actually be more rigorous 🔍✅, as he was trying to argue with the AI model 🤖💬.

He wrapped up by saying "that's it" 🎬, and then @nick-eagles.bsky.social enthusiastically responded that this is super useful 🌟 and they're probably going to switch over to using Positron today! 🚀😄

#RStats @posit.co

@geopertea.bsky.social discussed how the AI provides semantic understanding of code 🧠💻, showing how objects are generated from different parts of the codebase 🔗. He acknowledged that you can do this manually too 👨‍💻, and sometimes manual inspection might ...

#RStats #PositronIDE @posit.co

He pointed out - why not use AI to do that for you instead? 🤖💡 The AI quickly gave him the context of where it happens, functioning like a smart search tool ⚡🎯!

#RStats

He described another helpful scenario: tracing how a "log counts assay" object was generated 🔬. Normally, you'd search for the filename 🔎, then have to manually eyeball the context 👀 to see if that object was being read or written in a particular file 📄.

#RStats

@geopertea.bsky.social showed his chat history 📜 and explained a useful use case: he asked the AI to look into a repository 🔍 and tell him what methods they were using for calculations 🧮. Basically, he had the AI read the code and summarize it for him 📝✨.

#RStats #Positron #PositronAI

Once everyone agrees on the plan ✅, the idea is to switch to agent mode 🔄 so it has access to actually modify the code according to that plan 📝✨. That's the intended workflow! 🎯

#RStats #Positron

He explained the workflow: when you start modifying files or code 💻, it should switch to a different agent - either a specific agent or the main agent ⚡. The planning agent is good for creating plans through discussion 💬 between the LLM in the chat and the user 👥.

#RStats #Positron

@geopertea.bsky.social explained how you can add your own custom agents 🤖 to #Positron & showed the interactive way to configure them 🛠️. He demonstrated how the "plan" agent is defined, noting that it doesn't use any file mod tools 📝❌ - it's specifically designed for planning purposes 📋

#RStats

authentication code 🔑 which he pasted into the authorization page 🌐. He had to authorize the application to access his GitHub account ✅, going through two-factor authentication on his phone 📱. Once approved, he returned to Positron which confirmed the GitHub account was successfully connected! 🎉✨

@geopertea.bsky.social demonstrated setting up #Positron IDE 💻 by first detecting local interpreters and updating extensions 🔧. He showed how to login to @github.com to activate GitHub Copilot connectivity 🤖 for the Positron assistant. During the new installation, the app provides an...

#RStats

I explained why including context is crucial 📚! I pointed out that someone could just paste code like "GRanges from 1 to 1000 plus..." but then I would be confused 😕 asking "where does this function come from? What package is this from?" 📦❓

@jennybryan.bsky.social #RStats #reprex

I demonstrated running a #reprex in a fresh #RStats session 💻 & opened it in my browser 🌐. It showed the 📦 loading with all its messages, then the function itself. I pointed out that I had made a syntax error: I missed the colon before the plus sign ➕❌, which caused an error

@jennybryan.bsky.social

I explained how the reprex workflow helps with collaboration 🤝! I described a scenario where I ask a question about why the mean is 2.5 and not 3, and Nina wants to help by running the same code 💻. She can easily copy-paste the #reprex output and it actually runs ✅!

@jennybryan.bsky.social #RStats

I demonstrated how to use the #reprex package 📦! I loaded it and explained the workflow: I copied lines of code from my console (pretending Leo has a question ❓), then ran the reprex() function without any arguments 💻.

@jennybryan.bsky.social #RStats

I explained how a common problem makes it harder for people to get help 😤. People often think they've provided everything needed - posting their code online 💻, showing how to recreate objects, sharing all the code they ran, and even including the output 📊.

@jennybryan.bsky.social #RStats #reprex

I explained the common problem people encounter when asking for coding help 🤔. I showed how someone might copy code they ran, paste it, ask a question. I was setting up to explain what's wrong with this approach ❌ of sharing code for troubleshooting purposes💻 #RStats #reprex @jennybryan.bsky.social

YouTube video by Leonardo Collado Torres [2020-04-03] How to ask for R help by creating a small reproducible example (edited on 2026-01-27)

I re-edited this 2020 video using Adobe Premier Pro's features for removing filler words, pauses, and text-based editing. In total it went down from 20:53 min to 12:41. I hope that you get to (re-)enjoy it ^_^.

@jennybryan.bsky.social

#RStats #reprex #reproducibility #JHPCE

youtu.be/LfUFxHFhXTE

Lots has evolved since these R Stats sessions @libdrstats.bsky.social, but we touch on the hot topic of integrating LLMs into data science, mostly with Github Copilot, in a couple sessions--

Introductory look: youtu.be/xuCQZ5_5KB0
Reusable prompts: youtu.be/mlMl81xRI_M

#RStats #LLMs

Keri Martinowich, Ph.D., Chief Scientific Officer | The Lieber Institute for Brain Development | Research & Discovery Lieber Institute for Brain Development Appoints Keri Martinowich, Ph.D., as Chief Scientific Officer Baltimore, MD (January 6, 2025) — The Lieber Institute for Brain Development (LIBD) announced the a...

Honored to be appointed Chief Scientific Officer @lieberinstitute.bsky.social. LIBD is a uniquely collaborative place with extraordinary biological resources and a commitment to translation 🧠🔬🧪. Grateful to my colleagues and excited to help guide our next chapter!
www.libd.org/keri-martino...

PS You can see Cynthia's growth on @github.com by comparing the number of commits per year in 2022 (before joining @lieberinstitute.bsky.social), then the end of 2023, to 2025 (she spent nearly all year with us) 📈