Researchers from the FlyWire consortium have achieved a remarkable feat - mapping a complete connectome for the fruit fly! In a huge win for opens science, they’ve released the MaleCNS v1.0 connectome for others to use as well. Building this scientific achievement, people have decided to use the fly connectome for… more creative uses. As it turns out, you can map images as visual inputs by simulating neuronal activations, and tune activation thresholds on a couple of selected output neurons to make decisions about tasks. We’ve seen Stonkfly - a fly for trading stocks, a fly that can parallel park and yes, even Doomfly for playing Doom. Our lab performs tons of large-scale quality assurance (QA) on medical images. We’ve created tools like MASI QA for quickly running through tons of scans at a time to notice any anomalies - things like head motion during scans, failures in processing, or incorrect body parts that slipped into datasets. Inspired by what I’ve seen the fly do, I used Claude to prototype The Fly Who Does QA. How does this work? As with previous works, we simulate inputs to the visual system and simulate neuronal activations. Then, we take the sum of the firing rates of two neurons and treat them as a confidence score. To train the fly, we tune thresholds for the confidence score to classify the images based on a set of training data (here, 150 passing images and 150 failing images). After tuning those thresholds, we can apply the fly to new datasets for the same QA task! On a held-out test dataset with 150 passing images and 14 failing images, we found 93% sensitivity and 77% specifity. That’s not bad for a fly! I also had Claude put together a dashboard with an animation of the fly performing QA. It’s remarkable what we are able to do with AI agents these days. You can have a fly up and running tasks for a grad student in just a couple hours over the weekend! Ask not what you can do for the fly connectome, ask what the fly connectome can do for you.

> I’m starting a series called `/lost+found/net` to highlight corners of the web that deserve sharing. When I first applied to PhD programs, I emailed a professor who wanted a link to a personal website instead of a traditional CV. So, I put together this site. Personal websites for researchers have been around for decades, and it’s a great way to showcase your research independent of your institution. Today, though, our academic identities are at risk of being fragmented across different sites, publishers, and tools. There’s GitHub for tracking your open source code contributions, Web of Science for marking down peer reviews, and sites like Google Scholar and Scopus that index your publications. You can take control of your academic identity online using tools created by the IndieWeb, a community of people focused on making independent personal websites that give you control over your data. It’s decentralized - there’s no one way to be a part of the IndieWeb, but there are some common principles shared among the community. What can I do using the tools of the IndieWeb? I can respond to other people’s sites using webmentions, and their responses appear on my site. I syndicate posts from my website out to social media like BlueSky and Mastodon and receive likes and comments back on my site using Bridgy and BridgyFed. I’ve set up a way to post directly to my site from my phone using a Micropub server. To get started, check out the guide on the IndieWeb’s wiki, which is a handy resource. There’s plugins available for many IndieWeb tools to integrate into however you are creating and hosting your website. The IndieWeb IRC channel on #indieweb on irc.libera.chat (also bridged onto Discord and Slack) is welcoming and helpful. There’s room for innovation in bringing IndieWeb principles to researchers. Could we post preprints on our websites and automatically send them out to preprint servers like arXiv? What would it look like to automatically draw in papers indexed from Scopus onto our CVs? Can we share and respond to papers independent of any individual social media platform? A Google Scholar page doesn’t have to define you as a researcher - with the IndieWeb, give your work a home.

A map of part of a mouse brain, which is expected to be generalisable to people, could help scientists understand behaviours, consciousness and even what it means to be human

Import all tweets exported from X/Twitter to a Bluesky account. - marcomaroni-github/twitter-to-bluesky

Visitors to Nashville’s Adventure Science Center can get a new perspective of themselves with an augmented reality (AR) anatomy that creates the illusion of an X-ray view of the body’s muscles and bones.

Generate a surface to show the deformation from registering subjects to an atlas - MASILab/atlas_surface_rendering