Tony L. @tlint.bsky.social · 04/10/2026I’ve been looking for a a new drawing program. This one might be a winner for me. #compchemreshiki.comChemical drawing, reinvented.Chemical drawing, reinvented. Draw molecules, build reaction schemes, and prepare figures with ReShiki. 030
Tony L. @tlint.bsky.social · 11/09/2026This is probably the best one yetinpreparation.substack.comOpinion: I Miss When The NIH Payline Was 40% And We All Mouth-Pipetted PhenolBy Dr. Douglas R. Waverly, Professor Emeritus of Biochemistry, Yale University 001
Tony L. @tlint.bsky.social · 09/09/2026Recently updated my side project to render in @marimo.io notebooks! The project is built on #RDKit’s feature factory. The repo on GitHub: github.com/tlint101/pha... 020
Tony L. @tlint.bsky.social · 08/09/2026Happy to say that I finally got around to trying out Marimo notebooks. It renders in the blog post. There were a few kinks, but easy once I figured it out. tlint101.github.io/practice-in-...tlint101.github.ioMarimo Testing – Practice in Code 001
Tony L. @tlint.bsky.social · 01/09/2026Long story short, I use dumb local models and still do a lot of coding myself. But they have been good to get me starting more quickly. 010
Tony L. @tlint.bsky.social · 01/09/2026Finally put my thoughts down about how I use LLMs and local models. Things move fast and my workflow may change in 6 months. We’ll see. tlint101.github.io/practice-in-...tlint101.github.ioHow I use LLMs – Practice in Code 100
Tony L. @tlint.bsky.social · 01/09/2026The perspective also mentioned @openadmet.bsky.social who are looking to generate quality data for ML models in the field. They also have a competition for models targeting CYP going on right now. And their blog posts are quite informative. Highly recommend! 130
Tony L. @tlint.bsky.social · 30/08/2026It goes to show that these models do not solve problems if you yourself do not know what to do with them. Many want to use AI, not for a specific problem, but for the sake of it. That just doesn’t work. 000
Tony L. @tlint.bsky.social · 30/08/2026I do have a coworker who pays for access to SOTA models. I would argue their productivity was not boosted at all. One could argue that it had a detrimental impact on them. 100
Tony L. @tlint.bsky.social · 30/08/2026I would like to, but I can’t swing the subscription costs for access. This is despite pressure from my PI. But I feel if they want to boost my productivity with these tools, then the lab/grant should fund me instead for access. Otherwise I will continue with my current set up. 100
Tony L. @tlint.bsky.social · 30/08/2026I don’t have access to SOTA models. I run local models on a (puny) Mac Mini, so my workflow is entirely different. 100
Tony L. @tlint.bsky.social · 30/08/2026An interesting setup and use case of LLM in a workflow to build and compare different models by Pat Walters patwalters.github.io/Let-the-Agen...patwalters.github.ioLet the Agents Do the BenchmarkingHow I benchmarked three new chemistry foundation models on two ADMET datasets without cloning a single Git repository. 100
Tony L. @tlint.bsky.social · 14/08/2026Anyone else having trouble looking up articles after the ACS redesign? #chemistry #compchem 000
Tony L. @tlint.bsky.social · 12/08/2026AF can also, by doing all the work, trick people who do not specialize in protein folding into thinking it that easy across all protein domains. There is a lot of nuance in protein modeling many can miss. 020
Tony L. @tlint.bsky.social · 06/08/2026You mean you don’t look at the graphical abstract and then move on? 000
Tony L. @tlint.bsky.social · 02/08/2026I have been thinking about this a lot. I am seeing students using a lot of AI. AI has its uses. But in my experience, when I push back and question some of the methods or ideas, they often regurgitate the thinking of the models, if I can get a response at all. 120
Tony L. @tlint.bsky.social · 30/07/2026Very neat. I’ll share it with others in my lab for a try. Thanks! 👏 010
Tony L. @tlint.bsky.social · 19/05/2026The developer is looking for feedback on the UX for Druse (github.com/Vitruves/Druse) For anyone in #drugdiscovery, #virtualscreening, #cheminformatics, please give it a try and give them some feedback. thanks!github.comGitHub - Vitruves/Druse: Metal-accelerated molecular docking studio for macOSMetal-accelerated molecular docking studio for macOS - Vitruves/Druse 000
Tony L. @tlint.bsky.social · 19/05/2026For the past few days, I have been trying out Druse, a native MacOS app for molecular docking and structure-based #drugdiscovery. It utilizes Metal, so you can dock multiple molecules using Apple silicon GPU. You can find the package here: github.com/Vitruves/Druse 310
Tony L. @tlint.bsky.social · 06/04/2026Around this time, I found a post on LinkedIn. I wasn't the first to do this on MLX. Another package was released that does this too! And they recently added conformation generation to their code. All of this is a great way to speed up workflows on MacOS! github.com/guillaume-os... #Cheminformaticsgithub.comGitHub - guillaume-osmo/mlxmolkit: MLX molecules accelerationMLX molecules acceleration. Contribute to guillaume-osmo/mlxmolkit development by creating an account on GitHub. 100
Tony L. @tlint.bsky.social · 06/04/2026This chart can be a little misleading. As it measures the clustering time, not fingerprint generation. Luckily, you can use multiple CPUs to speed that process up. I used the default #RDKit method for that, with an example found on the RDKit blog: greglandrum.github.io/rdkit-blog/p...greglandrum.github.ioSome new ways to speed RDKit calls up using multiple threads – RDKit blogLet’s take advantage of modern CPUs 100
Tony L. @tlint.bsky.social · 06/04/2026Using apple silicon, I was able to speed up clustering molecules using Butina and KMeans. Of course, BitBIRCH continued to be the fastest (which is done entirely on CPUs). These are all great for plugging into current workflows. #Cheminformatics #compchem #RDKit #mlx 100
Tony L. @tlint.bsky.social · 06/04/2026Unfortunately for me, my main day-to-day work is on MacOS. That inspired me to try to make a version using MLX github.com/tlint101/MLX...github.comGitHub - tlint101/MLXMolCluster: Clustering molecules using Apple Silicon GPUClustering molecules using Apple Silicon GPU. Contribute to tlint101/MLXMolCluster development by creating an account on GitHub. 100
Tony L. @tlint.bsky.social · 06/04/2026Nvidia's nvMolKit leverages their GPU for some RDKit functions, like clustering, generating conformations, etc github.com/NVIDIA-Digit... #Cheminformatics #RDKitgithub.comGitHub - NVIDIA-Digital-Bio/nvMolKit: A high-performance, GPU-accelerated library for key computational chemistry tasks, such as molecular similarity, conformer generation, and geometry relaxation.A high-performance, GPU-accelerated library for key computational chemistry tasks, such as molecular similarity, conformer generation, and geometry relaxation. - NVIDIA-Digital-Bio/nvMolKit 100
Tony L. @tlint.bsky.social · 15/02/2026Fun fact, that’s the same Murcko from the Bemis-Murcko scaffold 000
Tony L. @tlint.bsky.social · 15/02/2026There are so many things to juggle in #DrugDiscovery that it is easy to forget some things - such as binding affinity. A nice perspective by Mark A. Murcko tries to remind us otherwise. pubs.acs.org/doi/10.1021/...pubs.acs.orgThe Affinity AdvantageDrug discovery is a complex, multiparameter optimization process. I argue that a greater emphasis on optimizing binding affinity will accelerate drug discovery. Note that “optimizing” is not always sy... 100
Tony L. @tlint.bsky.social · 29/01/2026I forgot to carry my decimals 🙈. It is 0.02 secs for BitBIRCH and 0.11 secs for Butina. 000
Tony L. @tlint.bsky.social · 29/01/2026I finally got around to testing BitBIRCH. TL;DR on a set of 1,000 molecules BitBIRCH did it in 2 Secs, while Butina took 11 Secs. #cheminformatics #python tlint101.github.io/practice-in-...tlint101.github.ioPractice in Code - Testing BitBIRCH-Lean 100
Tony L. @tlint.bsky.social · 24/01/2026But hey, at least they got an essay published in Nature. 🤷♂️ 000
Tony L. @tlint.bsky.social · 24/01/2026Outsourcing our critical thoughts and our communication skills to these bots for the sake of “efficiency” or “output” will only make our skills atrophy. No one, especially people able to nab prestigious PI positions, should let that happen. 120
Tony L. @tlint.bsky.social · 24/01/2026In this case, a hard delete of that data. If you are going to outsource your thinking to these chatbots, at the very least, download and run these models on a local server that you control. 100
Tony L. @tlint.bsky.social · 24/01/2026After four years, we should know how they work. I think all of us (or at least most of us) knows that when you interact with these program it does not stay on your computer, but goes to a server. A server that you do not control. Once there, they have the ability to do with what they will. 100
Tony L. @tlint.bsky.social · 24/01/2026We forget that these AI chatbots are tools. And for OpenAI and other companies offering these services, they have their uses and their drawbacks. 100
Tony L. @tlint.bsky.social · 24/01/2026As someone going through a PhD program myself, the one thing I learned to value is critical thinking and communication. I don’t think it is difficult to realize that outsourcing our one ability that differentiates us (thinking and communicating) from AI programs is a smart move. 100
Tony L. @tlint.bsky.social · 24/01/2026AI is an incredibly useful tool. But it is a tool. I use it. Millions (Billions?) of people use it. So in an age when everyone worries how AI is going to take our jobs, why would anyone, espeically a PI, outsource their job to AI? There is a lesson to this. nature.com/articles/d41586-025…nature.com 100
Tony L. @tlint.bsky.social · 24/01/2026If you are going to outsource your thinking to these chatbots, at the very least, download and run these models on a local server that you control. 000
Tony L. @tlint.bsky.social · 24/01/2026After four years, we should know how they work. I think all of us (or at least most of us) knows that when you interact with these program it does not stay on your computer, but goes to a server. A server that you do not control.Once there, they have the ability to do with what they will. 100
Tony L. @tlint.bsky.social · 24/01/2026We forget that these AI chatbots are tools. And for OpenAI and other companies offering these services, they have their uses and their drawbacks. 100
Tony L. @tlint.bsky.social · 24/01/2026As someone going through a PhD program myself, the one thing I learned to value is critical thinking and communication. I don’t think it is difficult to conclude that outsourcing our one ability that differentiates us (thinking and communicating) from AI programs is a smart move. 100
Tony L. @tlint.bsky.social · 07/11/2025A list of this year's and previous year's recipients can be found here: research.google/programs-and...research.googleGoogle PhD Fellowship recipientsThe Google PhD Fellowship Program recognizes outstanding graduate students doing exceptional work in computer science, related disciplines, or promising research areas. Meet the recipients. 000
Tony L. @tlint.bsky.social · 07/11/2025Happy to be a recipient of the 2025 Google PhD Fellowship! I learned a lot preparing my application. I wrote about it. Hopefully it helps others or others in preparing similar applications. tlint101.github.io/practice-in-...tlint101.github.ioPractice in Code - Applying for the Google PhD Fellowship 110
Tony L. @tlint.bsky.social · 23/07/2025I wrote about it in a blog post here: tlint101.github.io/practice-in-... (It is basically the tutorial section, but it counts!)tlint101.github.ioPractice in Code - Announcing the pharmacophore-toolkit! 111
Tony L. @tlint.bsky.social · 23/07/2025There are also neat 2D images that can be generated if you prefer. And the data is generated as a .pml file, which can be drag and dropped right into PyMOL to generate images like these. 100
Tony L. @tlint.bsky.social · 23/07/2025I wanted a tool to build simple pharmacophore models. It had to fit into my workflow and be able to generate fancy pictures. It includes a neat py3dmol jupyter widget. Perfect for those sharing Jupyter Notebooks! 100
Tony L. @tlint.bsky.social · 23/07/2025My first post and I would like to plug my project - pharmacophore-toolkit. I thought others in the #cheminformatics, #molecularmodeling, #drugdiscovery, or the #chemsky #compchem community would find it helpful. github.com/tlint101/pha...github.comGitHub - tlint101/pharmacophore-toolkit: Generate simple pharmacophore models from RDKitGenerate simple pharmacophore models from RDKit. Contribute to tlint101/pharmacophore-toolkit development by creating an account on GitHub. 1133
