Automatic Generation of Novel Highly-Functional Molecular Sequences

Vincent R. Ragusa; Chrisoph Adami

Home/Archive/arXiv

arXiv · 2611.03603

Automatic Generation of Novel Highly-Functional Molecular Sequences

Vincent R. Ragusa·Chrisoph Adami

View original source ↗Read source document ↗Export metadata ↓

Abstract

While multiple methods now exist to generate functional sequence variants from existing annotated examples, it is unclear how well such methods can be used to generate sequences that exceed the functionality of those contained in the example set. Using two complete fitness landscapes, we quantify how effectively generative tools can search sequence space and find rare, novel, highly functional sequences, focusing on variational auto-encoders and information-theoretic methods. We find that an information-theoretic method that does not require training and runs significantly faster than standard Machine-Learning/Artificial Intelligence (ML/AI) methods also is significantly more effective in finding rare highly-functional sequences. We discuss the application of this novel method to discover rare polypeptides with anti-bacterial properties that are specific to bacterial, but not eukaryotic, cell membranes.

Explore related subjects

q-bio.BM

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Vincent R. Ragusa, Chrisoph Adami. 2026-11-02. Automatic Generation of Novel Highly-Functional Molecular Sequences. https://arxiv.org/abs/2611.03603

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Predicting protein structures at any folding stage from sequence

Protein structure predictors reliably identify native conformations, but do not describe the continuum connecting unfolded ensembles to folded states. Here we present FoldScape, a sequence-conditioned deep learning framework that generates C$α$ conformational ensembles at a specified folding fraction $Q$. FoldScape reconstructs three-dimensional chains through hierarchical, sliding-window regressions of one-dimensional inter-residue distances, followed by chain assembly and sampling with a rectified-flow decoder. It is trained on multi-scale coarse-grained folding trajectories and all-atom mechanical-unfolding trajectories. Across 20 human proteins (36-857 residues), at matched sample size, FoldScape covers about an order of magnitude more of the folding coordinate than seed-varied AlphaFold 3 ensembles, robustly across proteins, order parameters and binning choices. AlphaFold 3 better anchors the immediate native vicinity, which FoldScape does not reach, making the approaches complementary. FoldScape reconstruction accuracy remains approximately constant along $Q$, from collapsed coil to near-native globule, rather than deteriorating away from the native state. A stride-one crop-and-superpose scheme extends inference to chains several times longer than a single chain-level forward pass. We also show that training trajectories resolve mechanical unfolding at residue resolution, that elongation-conditioned ensembles exhibit reproducible continuous observables across pulling velocities, and that partially unfolded chains form geometrically coherent segments. FoldScape thus combines broad coverage of folding stages with stable reconstruction accuracy, without per-protein simulation.

q-bio.BM↗

arXiv2026

Ablating Design Choices for Protein Ensemble Generators Across Structural Regimes

Generative models of protein conformational ensembles inherit much of their design from structure prediction for folded proteins. Ordered and disordered proteins, however, occupy distinct structural regimes, and we investigate how design choices hold up across them. Using a minimal C$α$ flow-matching model, we ablate choices of architecture, sequence conditioning, rigid-body alignment and training source, and evaluate each variant against experimental SAXS and NMR chemical-shift data partitioned into ordered, disordered, multi-domain and transitional proteins. Mixture-of-experts routing, a static pair representation and a disorder probe have no detectable effect, a smaller protein language model performs worse on ordered proteins but not on disordered ones, and alignment in the loss degrades agreement with SAXS. Replacing coarse-grained simulations of disordered proteins with fragment-assembled ensembles improves agreement with chemical shifts at a cost to SAXS, while an equal mixture of the two keeps most of the gain at a smaller cost. This points to the training data, rather than the architecture, as the main constraint.

q-bio.BM↗

arXiv2026

AI-driven structural informatics reveals diverse glycosylation machinery across double-stranded DNA viruses

Glycosylation is widespread in biology and shapes molecular recognition, yet the diversity and biosynthetic organization of viral glycosyltransferases (GTs) remain poorly understood. Here we integrate structural-alphabet-guided searches and structure similarity networks with genomic-neighbourhood and biosynthetic gene cluster embedding analyses to examine viral GT-A proteins. This approach reveals extensive structural diversity overlooked by sequence-based annotation, including previously uncharacterized viral GT lineages. Most strikingly, we identified a jumbo-phage locus encoding five tandem glycosyltransferases alongside hydroxymethyl-pyrimidine biosynthetic and sulfation-associated enzymes, suggesting a previously unrecognized form of viral DNA modification involving a sulfated pentasaccharide. Beyond these examples, predicted saccharide-associated loci span divergent double-stranded DNA viruses, from giant viruses to tailed phages, and occupy multiple regions of the reference biosynthetic landscape. Together, our findings uncover an underappreciated repertoire of virus-encoded glycosylation machinery and suggest that viral glycan biosynthesis encompasses diverse molecular contexts and potential biological functions.

q-bio.BM↗