Bio x AI 生命の特異点 / Singularitas biologica

Joined October 2024
63 Photos and videos
Diffusion language model speed is unreal. Here is DiffusionGemma vs Claude Haiku vs GPT 5.4-mini (low reasoning) racing to generate matplotlib figures. The models were prompted to generate Julia-set fractals, magnetic-pendulum basins, blackbody curves, Chladni plate modes, and hydrogen-orbital grids.
Introducing DiffusionGemma, our first exploration with open diffusion text generation models 🔥Generate blocks of text at a time 🤏26B MoE built on top of Gemma 4 ⚡️Up to 4x faster in popular consumer GPUs 🤗Apache 2.0 Excited to see what the community builds with it!
1
7
21
2,124
For testing, DiffusionGemma was run on an NVIDIA H100 through RunPod, served with vLLM, with weights from @huggingface (google/diffusiongemma-26B-A4B-it).
1
3
189
AI for redesigning molecular biology tools like affinity tags. Here's a short demo testing ESMFold2 to redesign the FLAG affinity tag using the anti-FLAG antibody structure (M2, PDB: 8RMO) as a reference. Target to redesign = FLAG: DYKDDDDK ESMFold2 was also tested in the opposite direction, where a new potential FLAG tag binder was designed. AI agents made this presentation for the demo using ChimeraX and the OpenAI text to speech API.
5
50
234
21,348
Also, an alternative potential FLAG tag binder was designed. In this example, ESMFold2 designed a 158-residue binder that recovers the six FLAG tag contacts the M2 Fab makes.
1
4
588
For this demo, ESMFold2 was run on Modal, a serverless GPU platform. Total cost was about $5 for ~200 designs.
2
6
523
An interesting constraint for theoretical binder design is limiting the amino acids that can be used. In this example, ESMFold2 was tasked with generating potential binders for GLP-1R using only: - 8 amino acids: SALTYGEK - 6 amino acids: YAKLEG - 3 amino acids: LEK The goal for the design was just to place candidate mini-proteins near the GLP-1R extracellular-domain groove where semaglutide/Ozempic contacts. Fun to see the structures that resulted! Codex put together this nice presentation of the results.
I'm so excited to show the world what we've been working on the for the past months!! I'm going to highlight some of the fun results from this paper that I find particularly exciting.
4
13
94
10,570
The potential binder when only the amino acids L E K are allowed ...
4
281
Here's an ESMFold2 demo run by AI agents on Modal for designing potential GLP-1 receptor binders. This test focused on the GLP-1R extracellular domain where semaglutide/Ozempic binds. It measured how well each design recovered the residues that semaglutide contacts. Codex made this neat presentation of the demo with autonomous use of ChimeraX and ElevenLabs.
I’m so excited about the launch of ESMFold2, ESMC, and the new ESM Atlas. This was a massive team effort, and I’m grateful to have worked with such an incredible group @biohub. A headline result I’m especially excited about: ESMFold2 can design minibinders and antibodies with nanomolar affinity, target selectivity, and functional activity against therapeutically relevant targets. Today, we’re sharing the full binder design protocol.
3
20
140
18,929
Codex for a mol bio suite and lab notebook would be nice
1
20
2,150
Grok Build with Composer 2.5 visualizing predicted binders against PD-L1, a cancer immune-checkpoint protein that tumors use to switch off T-cells. Composer 2.5 in Grok Build is using ChimeraX to visualize PD-L1 binders generated by Genie 3. In this workflow, a region on PD-L1 was selected for binding, Genie 3 generated binder shapes, ProteinMPNN filled in the amino acid sequences, and Boltz folded the binders against PD-L1.
Jun 1
We are excited for all of you to try out Composer 2.5 in Grok Build starting today! To use composer-2-5 do `/model` in Grok Build and type in Composer to switch Composer 2.5 comes with 200k context window and supports: subagents, MCPs, skills and additionally also works with your .cursor settings
8
53
7,375
AI for genome mining. Many high-value natural compounds are made by enzymes whose genes group into clusters. For many compounds the pathway is not fully elucidated. Finding those genes, and later testing them, helps piece a pathway together. For instance, a gene that is near a known anchor gene becomes a candidate for a missing step, even when the anchor is only an approximate match. As a demo, Claude picked berberine as a target compound, then ran bioinformatic tools across different plant lineages to compare the pathway's gene architecture and flag candidate genes for open steps. Claude also made this short presentation!
1
9
60
2,762
Claude searched four plants for gene clusters related to the berberine pathway. This included the berberine bridge enzyme (BBE), the CYP80 and CYP719 P450s, and O-methyltransferases. 1. Chinese goldthread (Coptis chinensis): the canonical berberine/BIA pathway reference 2. Fourstamen stephania (Stephania tetrandra; TCM han fang ji): known for coupled bisbenzylisoquinoline alkaloids like tetrandrine 3. Fish mint (Houttuynia cordata): where Claude found many BBE-like paralogs spread across the genome 4. Amur cork tree (Phellodendron amurense): a published P450 route, used as comparative context Here is an example where Claude found a tandem of three CYP80-like paralogs beside an anchor gene.
1
1
4
261
For researchers interested in using AI agents in their bioinformatic workflows related to genome mining and gene discovery, you can point your preferred agent (e.g. Claude Code, Codex) at this skill repo I put together. This repo provides an additional harness layer for structured gene cluster analysis, bioinformatic tool use, and comparative pathway mapping. If you are 'compute poor' like many researchers, this repo helps your agent use cloud/neocloud resources such as AWS, Runpod, and Lambda. It also nudges your AI agent to keep logs and learnings so your agent can adapt and get better at your workflows. github.com/BioSymphony/genec…
5
245
AI for structural biology: given the right harness, tools, and compute, your preferred AI agent can tackle tasks such as building a protein structure from a cryo-EM density map. In this case, Claude assembled a structure, compared its reconstruction to the published structure, then used ChimeraX to create these visuals and make this short presentation video.
6
44
254
16,604
Claude also compared several tools for building the protein from the cryo-EM density. In this case, Claude used ModelAngelo, CryoAtom, and Cryo2Struct, and then constructed this dashboard to display the results.
1
4
733
For structural biologists comfortable with tools like Claude Code or Codex, you can point your AI agent at this skill repo to give your agent a better harness for cryo-EM and related structural workflows. Also, a big issue with a lot of structural work is the amount of compute and storage needed. Neoclouds are great for this, and this repo points your agent at some of these resources. It also nudges your agent to set up self-learning / memory so your agentic workflows can improve over time. github.com/BioSymphony/cryoc…
12
662