The front page of AI-driven biology
Thursday, October 15, 2026Thu, Oct 15
Updated 7:31 PM PT
Computational biologists built OmegaTherm, a sequence-based language model framework designed to jointly estimate shifts in folding free energy and melting temperature caused by single or multiple mutations.
Scientists built Generative Attribution Metric Analysis (GAMA), an interpretability framework based on Integrated Gradients for autoregressive biological sequence generators.
Researchers used deep-learning design methods to transform the cancer driver K-Ras(G12C) into an inert 19-kilodalton self-labeling protein tag called LUCI-tag that rapidly binds commercial covalent inhibitors carrying fluorescent payloads.
Researchers used sequence- and structure-based computational models to design FlexID, an engineered ligase enzyme that labels neighboring proteins without releasing diffuse reactive tags.
Researchers used the generative protein-design model RFdiffusion alongside affinity maturation to create artificial proteins that selectively bind unanchored ubiquitin, an elusive signaling molecule lacking high-affinity probes.
Genentech is paying $55 million upfront in a multi-target oncology collaboration that could exceed $1.5 billion with development, regulatory, and commercial milestones.
Researchers developed a machine-learning method that generates target preference maps to estimate protein binding compatibility directly from native crystal structures.
Researchers have built Cerebra-Epistasis, a computational tool that combines single-sequence 3D protein structure prediction with an epistasis model to forecast how combinations of multiple mutations alter protein function.
A computational workflow predicts how effectively bacterial beta-lactamase resistance enzymes degrade carbapenem antibiotics using only the enzyme's sequence as input.
Researchers have developed PepSpace, a computational platform that pairs coarse-grained molecular dynamics simulations with a genetic algorithm to design peptide binders from scratch.
A new mathematical framework recasts spectral information theory into an interpretable metric called Leff, which calculates the effective number of fully independent positions that capture sequence variation.
Researchers applied the protein design algorithms PROSS and FuncLib to redesign human T cell receptor constant domains, creating a variant called structurally enhanced TCR that boosts surface expression and prevents faulty pairing.
Researchers used structural modeling from AlphaFold3 to engineer dual-excitation ratiometric biosensors that track neuropeptides in live animals without interference from blood flow, tissue movement, or pH shifts.
Target-adaptive Rescue-Explore-eXploit, or T-REX, uses large language model agents to orchestrate multiple generative models and structure-evaluation tools during protein binder design campaigns.
Researchers computationally designed synthetic transmembrane peptides that bind directly to the apolar, membrane-spanning domain of Toll-like receptor 4 to block inflammatory signaling.
Researchers combined the generative models ESM3, ProteinMPNN, and EvoDiff to design 76 functional carrier domains for non-ribosomal peptide synthetases, the giant multi-part enzymes that bacteria use to assemble complex peptide drugs.
Computational biologists created cytokine prodrugs designed from scratch that stay inactive until cleaved by tumor-specific enzymes.
Peptides computationally engineered purely to change their secondary structure in response to environmental cues proved to act as potent antimicrobial agents in lab tests.
Researchers introduced latent generative search, a framework that guides the Proteina-Complexa model at inference time to generate protein sequences and 3D shapes simultaneously, eliminating the need for a separate inverse-folding step.
Researchers combined machine-learning models of enzyme electrostatics with molecular mechanics and microkinetic modeling to simulate multistep catalytic reactions without relying on empirical approximations.
Bioengineers coupled the genomic foundation model gLM2 with discrete diffusion to guide the generative design of complex, multi-domain biosynthetic gene clusters.
Computational biologists generated thermostable, de novo mini-proteins that bind Nipah and Hendra virus surface glycoproteins with subnanomolar affinity.
Goa Therapeutics launched with $15.5 million in funding from individual industry leaders and filed for an initial public offering to raise up to $75 million.
A computational protein-design system combined sparse experimental measurements with evolutionary and structural data to simultaneously engineer catalytic efficiency, expression yield, and thermal stability in the industrial ketoreductase Gre2.
A wet-lab validation study comparing generative adversarial networks and variational autoencoders found that sampling protocols, rather than model architecture alone, dictate the experimental success of computer-designed antibodies.
A computational study demonstrates that combining machine-learned interatomic potentials with electrostatic embedding can accurately calculate catalytic reaction barriers in designed enzymes without costly quantum mechanical simulations.
A retrospective benchmark using Pfam updates revealed that structural homology searches using Foldseek recovered correct functions for only 15.9% of previously uncharacterized protein domains, compared with 30.1% for difficulty-matched known domains.
Computational biologists developed a multistate design framework that optimizes enzyme active sites across an entire reaction pathway rather than modeling a single rigid transition state.
Computational biologists introduced PANDA, an end-to-end generative architecture that designs proteins by denoising atomic coordinates in unified 3D space and reading sequence directly from atomic positions.
Researchers introduced Fluxion, a generative deep learning framework that models enzyme catalysis by generating dynamic, multi-step electron-flow pathways across active-site residues.