Among many things, in this episode we discuss:
- Obsidian's lead program, OBX-115, a modified TIL cell therapy
- The critical role of TCR sequencing in profiling infusion products, tracking clonotypes, and identifying biomarkers for therapy response.
Listen/View here:
youtu.be/J41q0fUqeVo?...
You can use add your OpenAI API key under the Advanced Setting to start annotating abstracts on your own. For those of you who don't feel comfortable with this, I have also decided to release the source code (feel free to reach out to your IT to help with the setup): github.com/armish/abstr...
For those of you who are trying to digest the recently released ESMO 2025 abstracts, abstract.plus now runs a web tool that helps you filter/annotate abstracts with a few clicks.
that level of accuracy (against multi-lingual text) will only come from LLMs and then the only light-weight solution would be making a call to a relatively cheap/fast model API (like gpt* nano). For production-level (hands-off) solutions, I would definitely take that route. It is worth the money.
This is meant for data scientists who don't use Windows hence suffer from embedded graphs due to limited OLE capabilities on non-Windows systems.
I would love to test this on as many slide decks as possible so any feedback is more than welcome!
Here is my second vibe coding experiment:
github.com/armish/prism...
a Python package (with a CLI) that can extract embedded @GraphPad PRISM plots from @powerpoint slides and insert them back.
Next step is to build a meaningful MCP server on top of this library but here is the {plumber2mcp} package for people who want to give it a try: github.com/armish/plumb...
Took only 2 days of intermittent collaboration with
@anthropicai.bsky.social 's Opus/Sonnet 4 to start with the idea and come up with a functional MCP solution.
I even let Claude Code handle the commits and most of the git plumbing -- and loved that @github.com captures this
Inspired by fastapi-mcp package, I (vibe-) coded {plumber2mcp}, a simple package to help add MCP support to a {plumber} API:
#rstats
pr('plumber.R') %>%
pr_mcp(transport='stdio') %>%
pr_run(port = 8000)
Supports HTTP/STDIO, tool/resource definitions, and schemas
Content re-distribution restrictions set by the organizers make it infeasible to release this as an open application but drop me a line (at arman@aksoy.org) if this is something you would like to beta test. I'd love to hear some feedback on its utility and learn more about its potential use cases.
In the video above, you can see me drilling down on bispecifics and extracting the details of the reagent via LLM's help.
With this app, it takes less than a minute to filter thousands of abstracts based on a keyword and just a few dollars to ask a question against those abstracts (via OpenAI's API). This is a simple web application that is easy to self-host and it only requires an API key entered into the settings.
It is so hard to browse, narrow, and annotate big conference (e.g. #ASCO25) abstracts using their original web sites. For the last 1.5 years, I have been experimenting with an AI-assisted way to consume these materials. I finally was able to combine my scripts into a single web application.
AlphaFold folks: is there a place where I can batch download MSAs (.a3m files) of the human proteome? Re-running mmseq2 on all the human proteins sounds like it would be a waste of time/resources?
Would have been convenient to have this companion GPT when I was doing my PhD 😭
Paper figure -> gene list -> comp'al friendly format with annotations
For those of you who want to play with an HGNC-savvy GPT for mundane gene ID mapping/look-up tasks, I present you Hugo the GPT: chat.openai.com/g/g-E7GrLSwn...
#AppleEvent 👏 Biomedical researchers 👏QuPath 👏 MinKNOW 👏
seconding Azenta. Pricier than Novogene but they are really great with getting the best out of problematic samples and providing customer support.
