C2Bio

Curated weekly digest of notable AI/ML and biotech developments.  This week's most exciting article may be RFDiffusion 3 paper from Baker's lab. Their preprint on bioRxiv was posted on November 19, 2025--so almost one year before it was accepted by Science.  https://www.science.org/doi/10.1126/science.aeh2982 Original preprint:  https://www.biorxiv.org/content/10.1101/2025.09.18.676967v2.full.pdf Some other paper interesting papers:  ★ A high-resolution protein-RNA binding atlas for structure-aware interaction prediction and post-transcriptional regulatory modeling — RNA-binding proteins (RBPs) control post-transcriptional gene expression by regulating RNA processing, localization, stability, and translation. ★ Mutational constraints on RSV F and its neutralization by antibodies — New antibodies targeting the F protein of respiratory syncytial virus (RSV) have substantially reduced infant hospitalizations 1 . ★ BioMod: Biological Module Tokenization for S...

You draw a figure with Python using matplotlib in PDF, open it in Illustrator to fix one label before the submission deadline, and the text is there but you cannot select it. Click a word and you get a cluster of tiny filled shapes that happen to be letter-shaped. The glyphs are right. The text is gone. This is not an Illustrator problem and it is not a corrupted file. It is a matplotlib default. Tested on matplotlib 3.10.8. Backend defaults do occasionally move, so check yours with matplotlib.__version__ if the behavior below does not match. Type 3 versus Type 42 The PDF specification supports several ways to embed a font. Matplotlib’s PDF backend defaults to Type 3, which defines each glyph as a small PostScript drawing procedure. A Type 3 font is a set of instructions for painting shapes. It carries no reliable mapping back to characters, so a design tool can render it faithfully and still have nothing to hand you when you click. Type 42 wraps a real TrueType font, outlines an...

🧬 Biotech ★ GraPPI: A Self-Supervised Graph Encoder for Transferable Protein–Protein Interaction Modeling — Protein–Protein Interactions (PPIs) are fundamental to cellular processes and represent major targets for therapeutic development. ★ High-throughput metabolic engineering in mammalian cells — The paper introduces a multiplexed shotgun engineering strategy that screens millions of combinatorial genomic solutions to engineer complex phenotypes efficiently in mammalian cells.  ★ What deep learning can and cannot (yet) do for enzyme design — Deep learning has substantially advanced structure-based enzyme design, yet reliably creating highly efficient enzymes remains out of reach. ★ Mapping transcriptional responses to cellular perturbations with RNA fingerprinting — A statistical framework that infers causal regulators from single-cell experiments by mapping their transcriptional responses onto reference perturbation dictionaries.  ★ AI-integrated RNA ...

Curated weekly digest of notable AI/ML and biotech developments. Flagged items are manually selected; remaining items are the week's most recent. The most interesting article from the past week is the " Function-preserving watermarking of AI-generated proteins" from Stutz et al. Their approach integrates watermarking into ProteinMPNN in the sequence design step on top of a backbone structure. Although I haven't tested it, this approach suggests generating specific patterns in amino acids used to fit specific backbone structure requirements. In other words, the sequence one can get through SynthIDBio (their methods reported in this paper) is a subset of all the possible sequences that can likely fold into the targeted structure. The sequence itself can be checked for patterns that suggests its origin. Similar methods have been used to watermark LLM generated texts. The ML realization part is simple and effective. Figure 1 in this paper showed in the Sampling phase the ...

Curated weekly digest of notable AI/ML and biotech developments. Flagged items are manually selected; remaining items are the week's most recent. 🧬 Biotech ★ SurfGraphPro: Integrating Protein Language Models with Geometric Deep Learning On Coarse Protein Surfaces for Binding Site Prediction — This work from my team introduced a new protein representation algorithm that bridges proteins structure model and language model. Our algorithm does not compute amino acid features computationally. Instead, it utilizes PLM embeddings plus protien surface contour coarse-grained graph model to replace the feature engineering. In test application focusing on protein binding interface prediction, we achieved state-of-the-art performance at a much faster (~18x-28x) speed. ★ GitHub - PacesaLab/BindCraft2: Easy to use suite for designing a wide range of protein binder modalities. — Easy to use suite for designing a wide range of protein binder modalities. ★ The Virtual Biotech: A mult...

Curated weekly digest of notable AI/ML and biotech developments. Flagged items are manually selected; remaining items are the week's most recent. 🧬 Biotech ★ QO‑BRA: A Quantum Operator-Based Autoencoder for De Novo Molecular Design — We introduce a variational quantum autoencoder tailored for de novo molecular design, named QO-BRA (Quantum Operator-Based Real Amplitude autoencoder). QO-BRA leverages quantum circuits for real-amplitude encoding and the SWAP test to estimate reconstruction and latent-space regularization errors during back-propagation. Adjoint encoder and decoder operators enable unitary transformations and a generative process that ensures accurate reconstruction, as well as the novelty, uniqueness, and validity of the generated samples. We showcase the capabilities of QO-BRA as applied to the de novo design of Ca2+-, Mg2+-, and Zn2+-binding metalloproteins after training the generative model with a modest data set. ★ The quantum ensemble variational opti...