Dr. Ashfaq Rehman, Ph.D., Author, Biophysicist @ocdrash

a nice demonstration of Claude Science, but worth clarifying that the design is not "done by Claude" but by orchestrating tool calls of open-source, task-specific protein design models: PXDesign, RFdiffusion, Genie, BoltzGen, etc I think the direction of LLMs using biology-specific models is a good one!

Anthropic @AnthropicAI

a month ago

Many drugs work by binding to a specific target in the body and blocking or changing what it does. An important first step in the drug development process is designing a molecule that can bind tightly to its target. Traditionally, that's meant weeks or months of expert work per

606 1K 12K 4.3M 4K

New pub from @diffUSEproject! Ordered water molecules mediate protein stability, ligand binding, and catalysis. Predicting them has lagged behind protein structure prediction. We developed WaterFlow, a new state-of-the-art method for placing ordered water. biorxiv.org/content/10.648…

Some of the hardest proteins to drug are difficult precisely because they refuse to hold still. α-Synuclein, the protein strongly associated with Parkinson’s disease pathology, is largely intrinsically disordered. One of its most aggregation-prone regions, the NAC domain does not present the kind of stable pocket that conventional drug discovery likes to see. A new bioRxiv preprint takes a different approach. Instead of searching for a pre-existing pocket, the researchers designed compact de novo proteins around a structural tendency of α-synuclein itself: the NAC region’s ability to adopt an extended β-strand during aggregation. The resulting binders recognize and stabilize this otherwise transient state. Then comes the more interesting part. The binders could strongly delay α-synuclein aggregation even when present at only around 10% of the α-synuclein concentration. That means they cannot simply work by capturing every monomeric protein molecule. Instead, they appear to target specific steps and species within the aggregation pathway. One design, 17_2, preferentially bound α-synuclein oligomers and selectively suppressed secondary nucleation, the process by which existing aggregates catalyze the formation of additional oligomeric species. Other binders also interfered with fibril elongation. This changes the therapeutic intuition. Instead of asking: “How do we bind the most abundant form of the protein?” we can ask: “Which rare molecular state actually drives the pathological process?” There is still a large gap from an in-vitro aggregation inhibitor to a Parkinson’s therapy: brain delivery, intracellular stability, immunogenicity, toxicity and efficacy in disease models remain unresolved. But the conceptual direction is powerful. An intrinsically disordered protein may not need a permanent drug pocket if we can design a binder for the fleeting structure that matters most.

keyboard_arrow_left Previous keyboard_arrow_right Next

Tell me you hate wasting time Initially took me hours👇 Now just 15minutes limewire.com/d/QbhcP#icWxiF…

A unified framework for molecular property prediction based on hierarchical multi-granularity molecular representation learning 1. The paper presents HMG-MRL, a unified molecular property prediction framework that explicitly integrates three complementary granularities: (i) fine-grained atom–bond structure, (ii) medium-grained functional motifs, and (iii) coarse-grained global molecular attributes (RDKit descriptors), aiming to capture chemical semantics that single-view models often miss. 2. A key design is an atom–bond bipartite graph encoder that treats atoms and bonds as two explicit, learnable node types in one message-passing system—so atom–atom, atom–bond, and bond–bond interactions are modeled jointly within each layer, rather than being handled indirectly or in separated pipelines. 3. For motif-level modeling, the framework builds a diverse motif vocabulary per molecule by combining three decomposition strategies: functional group matching, BRICS fragmentation, and Bemis–Murcko scaffold decomposition. Overlapping and nested motifs are allowed (a single atom/bond can belong to multiple motifs), preserving multiple valid “substructure perspectives.” 4. The motif view is encoded by a Motif Transformer: motif tokens are first made “atom-aware” via cross-attention that pulls context from the learned atom embeddings, then a Transformer with self-attention models global interactions among motifs; a CLS token summarizes motif-level information into a molecule representation. 5. A Global Attribute Encoder incorporates standardized RDKit molecular descriptors (physicochemical priors). It produces an attribute query vector and uses cross-attention to retrieve relevant context from atom embeddings, yielding an attribute-aware molecular representation rather than simply concatenating descriptors at the end. 6. Cross-granularity communication is built into the encoding process (motifs and descriptors attend to atom embeddings), and a cross-granularity alignment loss (contrastive objective) encourages semantic consistency between the atom–bond representation and the motif representation for the same molecule. 7. The final prediction uses attention-based fusion over the three granularity-specific representations (atom–bond Hb, motif Hm, descriptor Hd), learning per-task weights to combine them into a unified molecular embedding for classification (ROC-AUC) or regression (RMSE). 8. Evaluation on nine MoleculeNet benchmarks with scaffold splits (8:1:1) and five seeds shows strong overall performance: best mean results on 8/9 datasets. Reported highlights include ROC-AUC gains on BACE (0.910), BBBP (0.953), ClinTox (0.947), and improved RMSE on ESOL (0.690) and FreeSolv (1.541); Lipophilicity is slightly behind the top baseline. 9. Ablations indicate the gains rely on synergy: removing the bipartite encoder, Motif Transformer, descriptors, cross-granularity communication, or alignment loss all degrades results; descriptor removal hurts regression most, while motif removal impacts classification and regression, supporting the need for both functional substructures and physicochemical priors. 10. Interpretability analyses (t-SNE + clustering on BBBP; attention visualizations and attention-guided masking on FreeSolv/BACE/ClinTox) suggest the model’s attention highlights chemically plausible regions (e.g., polar groups for solvation, nitrogen-containing cores for BACE activity, thiophene/protonatable motifs for toxicity), and masking high-attention atoms/motifs harms performance more than random masking. 💻Code: github.com/AKZstar/HMG-MRL 📜Paper: doi.org/10.1093/bioinf… #ComputationalBiology #Cheminformatics #MolecularRepresentation #GraphNeuralNetworks #Transformers #DrugDiscovery #ADMET #MoleculeNet

Scientists unveil more than 600 new tissue models of human cancer: Derived from patient tumor samples and available to researchers around the world, the cells will aid the development of new cancer treatments. news.mit.edu/2026/scientist…

Since the landmark decoding of the human genome in the early 2000s, DNA sequencing has exploded. Traditional computers have struggled to keep pace with the deluge of data and soaring processing demands, creating a bottleneck in scientists’ capacity to mine the myriad variations in DNA for biological insights—and a push for alternative solutions. Now, one option, quantum computing, may be a step closer to helping. Researchers say they have for the first time encoded a complete, albeit small, genome, that of the hepatitis D virus, into a quantum computer, proving in principle these weird machines could one day aid genomics research. Learn more: scim.ag/4vAUeI0

What if designing a protein felt less like writing code and more like sculpting something with your hands? A new bioRxiv preprint introduces ProteinSketch, a VR environment where researchers can physically sketch protein backbones in 3D, shape the volume they want a protein to occupy, and let a diffusion model turn those gestures into molecular structures. Put on a headset, use both hands to draw a topology or define a curved 3D envelope, and RFdiffusion continuously refines that spatial idea into candidate proteins. This sounds almost like a futuristic interface demo, but the deeper problem is real: generative protein models are becoming extremely capable, while communicating high-level spatial intent to them remains surprisingly difficult. A sequence prompt is easy to type. “Build me a protein that bends around this molecular surface, enters this gap, and presents a binding interface here” is much harder. ProteinSketch makes geometry itself part of the conversation. The researchers used volumetric constraints to generate unusual anisometric protein shapes and confirmed shape fidelity by cryo-EM. They also used the same framework for functional binder design and to extend existing minibinders toward molecular surfaces that conventional design approaches struggled to reach. What I like about this direction is that it changes the interface to generative biology. We have spent enormous effort teaching models the geometry of proteins. The next challenge may be giving humans a more natural way to express the geometry they actually want. Maybe the future protein-design language will not be text at all. It may be shape, motion and space. Preprint : biorxiv.org/content/10.648…

keyboard_arrow_left Previous keyboard_arrow_right Next

It’s was very interesting discussion about what job, you don’t want even if the pay is 1million. Unfortunately the post, I don’t know where is violating. Reddit deleted it. reddit.com/r/whatdoIdo/s/…

deepseek v4 flash designing a synthetic protein ring and validated with boltz-2 fold and gromacs simulation

keyboard_arrow_left Previous keyboard_arrow_right Next

Today I'm announcing a protein compute web service: subseq.bio It's a simple web tool to submit compute-intensive tasks related to protein design, and biotechnology jobs. You can select from state-of-the-art, open source protein ML models, pre-configured and can begin running within minutes. Currently supported are RFdiffusion, and Alphafold (v2), with more coming. There are no restrictions on your output data. Just select the program, enter program arguments, and an optional upload folder, and click submit. Download the output data when finished. It is pay-per-use: Jobs are charged from your account credits when submitted. You can top up credits easily, any time. Currently jobs cost between 5¢ - $1. Feedback is important! Let me know ways it can be improved. More features are coming. For detailed updates please follow @subseqbio

Get out your calendars, because the new 2027 EMBL Annual Poster has arrived: s.embl.org/2027-poster Whether you'd like to enhance your skills in a practical course or engage with the latest research at one of our conferences, we've got you covered. See you at @embl in 2027! 🙌 #EMBL #EMBLEvents #LifeSciences #LifeScienceTraining #MolecularBiology

im not even gonna post the photo man

The New York Times @nytimes

2 months ago

Breaking News: Scientists have used A.I. to create new viruses for the first time, raising hopes for medical advances while also raising the possibility that the technology could someday be used to invent dangerous pathogens. nyti.ms/4bxx7Wy

715 1K 4K 4.2M 2K

@NewsfromScience Lovely discovery

Physicists say they have successfully demonstrated via high-resolution simulations that a DNA molecule can in principle behave like an Archimedes screw in nanoscale. Learn more: scim.ag/3TerI0d

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…

@elonmusk Any space for passionate and oriented individual with convincing potential in SALE of analytical and biochemical, weather SpaceX interested to allocated those top tier? Compt. biophysicist

Direct protein sequencing via nanopores is hard. Peptides move fast. So glue the protein to the pore so it can't move + use an enzyme to cleave off one amino acid at a time. Each amino acid can then be measured 100s of times. They sequenced a 22 a.a. protein with 98% accuracy!

Huang_Nanopore @Huang_Nanopore

2 months ago

New Huang lab publication at Nature. Our newly developed technique: transient Pore Analyte Looping (tPAL) now achieves “chop and measure” nanopore sequencing of peptide, in an amino acid by amino acid manner. nature.com/articles/s4158…

16 115 552 83K 239

@benjitaylor @elonmusk Is it possible to issue free for a month for expert level candidate?