##Tools

###Play Overseas YouTube Videos in Japanese Audio

It just lowers the original audio volume, gets subtitles, and reads them out, but it seems useful. Since it's a personally developed tool, it might suddenly stop working.

For some reason, pasting URLs didn't find videos, but searching by words worked for video search.

I've been practicing English daily since I use it a lot for work now, but if I could listen in Japanese, it would make passive input possible, which is nice.

###Claude3

What else could it be this month? Claude3 Opus is just godly. Especially the combination with perplexity.ai is amazing - the AI experience provided by web search and an LLM better than GPT-4 is the best ever. I revived my subscription that I had for just one month during Claude 2.0. (But then I decided to use perplexity instead, so I cancelled Claude and started subscribing to perplexity)

Particularly triggered by this tweet, I also seriously got into Amazon Bedrock. I'm actually using it for work too. The minimal model Haiku is also much smarter than GPT-3.5, so for casual uses you can create very cheap, fast-responding applications. Highly recommended.

###Arc browser

A browser I had been curious about for a while. The Windows version invitation email finally came so I can use it now.

There are quite a few noticeable bugs, but like SideKick, being able to pin frequently opened pages on the left side is incredibly convenient. Also, new tabs can be opened like spotlight, so being able to jump straight to search from Ctrl+T is really nice.

I was using SideKick before Arc browser, but that required a subscription for the full version, and it automatically added Gmail signatures, which felt spammy.

###Obsidian Thino

After wandering in search of the best editor for several months, I've finally settled on Obsidian these days. I learned about Obsidian Thino, a daily note-taking extension, and have been happily taking notes.

With Thino, you can take notes with the feeling of tweeting posting on X, and daily summary files are automatically generated. I had been struggling with how to take lab notes for bioinformatics work until now, but I think this is the optimal solution (at least for now). I'm trying to take notes on even trivial things.

##Python Libraries

###img2table

An OpenCV-based library. Can convert from images to tables. It depends on standard OCR libraries like TesseractOCR, but it's great that it converts cleanly to pd.DataFrame in one go. Looking at the analysis examples, accuracy seems quite good. Will try it when I get the chance.

###BASALT

A binning tool refinement tool. I don't plan to do metagenome analysis anytime soon, but this kind of tool is valuable so noted.

Paper

###Rworkflows

Can create Docker containers that define R version, dependencies, and LaTeX together.

Paper

###shu

High-dimensional signal pathway drawing tool. Looks easy to write in Python.

Paper

###Trackplot

I don't have a use for it, but an R script for creating epigenome heatmaps.

##Presentations

###Current State of LLMs

An excellent resource made by Preferred. I didn't know BERT-type models have better accuracy than LLMs for classification tasks (p.21).

###The Impact of ChatGPT

A summary of what you can do with ChatGPT, created by someone at Microsoft. Written in an engineer-friendly way, it was very helpful with content like "GPT-4 usage tips for beginners."

I've been doing a bit of RAG for work, and just seeing various use cases is eye-opening. Grateful.

###[Summary of Virtual Environment Management (env-type) Tools for Python, R, Julia]

A presentation by the reliable Professor Tsuyuzaki. I thought pyenv was fine for Python, and from a user perspective maybe so, but I learned that from a developer perspective that's not necessarily the case.

###9 Years of AWS: Pitfalls and Pain Points

Excellent. Perfect content for an AWS beginner like me who doesn't have time to seriously study AWS.

I've made the mistake of confusing terminate and stop once...

Video available too.

##Readings

##Papers

###Gut Bacteria's Indole Propionic Acid Enhances T Cell Stemness and Strengthens Immune Response

A mouse experiment. A common paper seeking gut bacteria's role in immune checkpoint inhibitor sensitivity, but with high resolution identifying even the metabolites.

Lactobacillus johsonil produces IPA (indole-3-propionic acid) through tryptophan metabolic pathway. IPA enhances acetylation of Tcf7's super enhancer region in T-cells, activating progenitor exhausted CD8+ T-cells (Tpex). It's interesting that Clostridium sporogenes also cooperates in IPA production, which is characteristic of gut bacteria.

The effect of these two species enhancing ICI was demonstrated through fecal transplant and bacterial transplant. IPA was added to CD8+ T-cells in vitro to observe T-cell activation.

###Fas-L/Fas Reduces CAR-NK, CAR-T Persistence

While the common knowledge is that Fas acts positively for cancer cells, single cell analysis of cancer patients found that Fas is highly expressed in T and NK cells while expression is low in cancer cells and stroma. Indeed, looking at CAR-T cells in patients who received CAR-T therapy, Fas-L was highly expressed.

What does this mean? Comparing Fas-L with anti-CD3/CD28 antibodies or antigen stimulation showed CAR-T/NK Fas-L expression increased in response to antigen stimulation. In other words, Fas-L expression increases in response to CAR-NK/T activation.

Since the Fas signal is an apoptosis pathway, using Fas KO CAR-T/NK to verify survival showed increased persistence. However, killing activity wasn't affected by Fas expression.

Up to here, we can say Fas signaling is a hot pathway involved in CAR-T/NK persistence. The CAR-NK boom has completely passed, but if this finding is true, it seems likely to be useful somewhere.

###Stem Cells with Multipotency Obtained from Amniotic Fluid

Simply amazing. There's risk of miscarriage with amniocentesis, but I've heard it depends on the obstetrician's skill. If you can get stem cells from this, you can get good cells with minimal risk. Wow.

##Blogs

###Continued: 4.5SH is Sweet (4) Light and Wandering at the End of Persistence

A blog post reflecting on the reality of research. I've never done transgenic mouse experiments, but I see, so knockout mice basically have no visible phenotype changes. The RNA-seq data analysis part was also educational in a way where I couldn't say "I could have reached that result faster if I did it" - learned something. Indeed, introns are interesting. Looking forward to the next post.

###Pharmaceutical Patent Cases

A blog that's been running for 17 years, but I just learned about it. I read some articles about CRISPR-Cas9, and all the judgments are well summarized and easy to understand. I wonder if it's mostly med-chem? It's interesting so I'll keep reading.

###(Finally) Trying the Gemini API

Gemini was offering a free API...? I'll use it next time I have an LLM need.

###Statistics Practice Workbook Complete Walkthrough (Chapters 1-4)

A Qiita article. People explaining workbooks are quite valuable. I haven't been studying statistics lately, but I want to seriously study for exams...

###DataArts Statistics Certification Level 1/Semi-1 Preparation Site

I used to study a lot using sites like this when I was at my previous job, but it's hard to find time at my current job...

###Cell Culture "Cell Count Measurement of Adherent Cells"

A story about how cell characteristics change based on time from dispersing ES cells to seeding. How results change when measuring cell counts with hemocytometers vs Coulter counters.

These verification articles are helpful and interesting.

##AI

###How to Write Meeting Minutes with the Highest Accuracy [ZOOM + Claude3 Opus]

This transcription method for Zoom was eye-opening. I was really surprised. The attached prompt and GPTs are also amazingly accurate, shocking. Making meeting minutes for Zoom meetings will be so much easier going forward.

###Hakky Handbook - Portal Site for Sharing Knowledge on Data and AI: LangChain Introduction and Use Cases Summary

Very valuable as there aren't many sites summarizing how to use LangChain in Japanese.

You might think "there are tons of articles on Qiita," and you're right, but those aren't quality you can reference.

I only started understanding LangChain after buying and reading a book, but honestly AI environments change daily, and I feel you can't truly master it without continuously keeping up.

Also, my AI use cases have been changing lately, and I use perplexity online far more than utilizing LangChain.

###Technique Collection for Improving Answer Accuracy with RAG

Applied Edition-A article came out recently.

I can't follow the latest papers in this field, so summaries like this are helpful.

##DB

###UCSC Xena

A web app that lets you quickly view heatmaps across various gene expression datasets. You can view correspondence with metadata in each dataset, so you can see gene expression by gender etc.

Since there's a ton of data in there, it might be useful for something if used well. But for now it's stuck at "a tool that creates cool-looking figures instantly and makes you feel fancy"...

###Genes Expressed in HEK293T Cells

This post was going viral (in the biology community).

###How to use Ensembl

Something I always end up searching for, explained casually in an X thread by the official account. Grateful.

###Azimuth App for Reference-Based Single-Cell Analysis

I learned about this from a recent common paper about using GPT-4 for scRNA-seq data cell type classification. Seems to be an example of manual cell annotation. You can never have too many datasets like this as an scRNA-seq analyst.

##Other

###Time-Saving Shortcuts Every Smart Colleague Knows

Excel, which I still use quite often. Logomix uses spreadsheets quite a bit too, and everyone uses Excel shortcuts quite a lot, so I read it. I didn't know about Ctrl+d.

###What Even Is AWS Lambda - ZennBook

Only read a little but seems good as an entry article. Will read later (a month has passed since I said that).

##Repositories

##AWS

###AWS Samples

A solution architect from Amazon introduced us to genomics-cli and Amazon Health Omics. Genomics-cli and SageMaker seem good for our use cases, but we haven't reviewed them yet.

###AWS Black Belt Online Seminar AWS Batch

Looks useful. I wonder if we could do job management with AWS Batch.