Rob Beynon

One for the scientists
My site for calculation of recipes for thermodynamically correct pH buffers has data about the buffers that are made.
phbuffers.org/BuffferCalc/...
Analysis of about 40,000 buffers. -> lot of the five most common buffer species. Anyone care to guess what these top 5 are?

As a welshman, I appreciate the Dylan Thomas, albeit mutated :)

Many of you (say, about one million) have used my buffer site to calculate thermodynamically correct buffers. Well, today, I upgraded the site with added security, becuase it was stopping soem people connecting. Should all be fine now. https rather than http. Enjoy!
phbuffers.org/BuffferCalc/...

Just before the Beynfest! Claude continues to impress. I asked it to write a simple protein dotplot tool to explain to a PhD student how two sequences are related, and 20 minutes later:
phbuffers.org/Claude/DotPl...
Free for all to use...

Slight ambarassed to be at the ’Putting Beynon out to pasture’ celebration at @ukbspr.bsky.social tomorrow. Thanks to the organisers…

I asked Claude to writ ethe annotation guide, and it built in features I did not ask for, and did not know about! I hope this will actually be of use..

phbuffers.org/Claude/pepti...

yes, me too. frequent saves, and 'start over' with that version. The dialogue with Claude, from the tool, actually talked about protein chemistry! I, and I suspect you, and the future generation will learn to give clear instructions.

Here's the annotation language for peptidemapper
phbuffers.org/Claude/pepti...

BTW, 1600 lines of code. SO impressed
phbuffers.org/Claude/pepti...

"I Claudiused" - tragic tale.
My tool is now silly. (haven't quite made a mess-alina of it though!). Has a flexible annotation language, which was not easy to get right. But "P14 sequence blue out above L1" means "Label P14 with its sequence in blue outside the map" @catfranco.bsky.social

Oh yes, the URL of the peptide mapper:
phbuffers.org/Claude/pepti...

So, I rewrote my little peptide mapping tool

And, in conversation with @catfranco.bsky.social she suggested a feature - so, in 5 minutes, it was implemented.with an annotation language

K fill cyan
C circle yellow in
V diamond green out below
P45 fill orange

Fun!

And here it is. OK, there is no doubt that writing the original taught me how to specify the task, and there is less functionality in some areas, but more in others, yet, within about 6h, here it is! phbuffers.org/Claude/pepti...

I wrote a neat little tool to draw peptide maps a long time (In Livecode/Hypercard) and in 4h have had Claude create a near identical web app...when it works well I'll share a hyperlink. @catfranco.bsky.social

I asked it to ‘prettify’ my buffer program.. made a total mess of it over and over again…so not perfect

Neat tool though… and it can always be improved . In fact hard to leave alone..

Loved it.. now you know where ‘MEERCAT’ came from…

Loved it! We are still looking for marilin- watch this space! Fantastic interview! Will see Kathryn in the BSPR meeting in Liverpool in July, www.bspr.org/annual-meeti... when there is a ‘putting Beynon out to pasture’ session! Payback for the embarassment!

neat - screenshot?

"vibe coding" 🤮 😂

capsaicin? Cathepsin is a cysteine proteinase e.g. papain, bromelain, so maybe in papaya and pineapple? :)

.

Any other stories like this? end/ Anyone used Ai to code scientific applications? I do think the main limitation is in specifying the parameters of the tools that I want to make. That is on me. end/

But, for ordinary, inept folk like me, this makes simple tasks completely within reach. e.g. I asked Claude for an SVG viewer for my peptide mapper programme -easy. Of course, I will now rewrite the entire tool as a web app.. so much more user friendly. 8/
(phbuffers.org/Research/Pma...)

Scarily, Claude not only correctly interpreted the Python code but also intimated that if I had the exe file, it could probably reverse engineer that too. 7/

And bingo, a web app that replaces an exe file with a fast and pretty, user-friendly flexible web app. I did no coding at all, and it took me less than a day of intermittent access. The decoding is instantaneous for large files. 6/

Breakthrough. A very clever research in our group had previously decoded the data, and we found the old Python code to do that. Showed this to Claude and bingo! Immediate result. After that, long dialogue about making it pretty, export options, tooltips and an explanation of the coding method 5/

So, Claude tried to build a lookup table to reverse engineer the time from known examples. Many days of data, at very high time resolution, was never going to happen, as it required just too many data points, and even then, no guarantee all possibilities would be covered. 4/

But we hit an interesting sticking point. Claude correctly identified which bytes encoded the timebase, but could not unpack the highly compressed format (it involved a variable number of sixteenths of a second increments) 3/

The first attempt was poor, and with a bit of inspection, I was able to let Claude know what the EOF and EOR markers were, as well as indication what the columns of data were when unpacked. Second attempt - so much better. 2/