Biology subjects

Ferruz, N.

Publications and source records attributed to Ferruz, N..

11 recordsLinked to original sources

bioRxiv2026

Translating chemical reactions into enzyme sequences enables formate dehydrogenase engineering

Enzyme engineering has broad applications across chemistry and biotechnology but remains costly and susceptible to low success rates. Generative artificial intelligence can substantially expand the sequence space accessible to enzyme engineering; however, directing generation towards enzymes that catalyse a user-defined chemical reaction remains challenging. Inspired by machine translation, we formulate enzyme generation as translation between two molecular languages: chemical reactions and amino acid sequences. Here, we introduce REXzyme, a translation model that generates enzymes conditioned on user-defined reactions. First, we evaluate its designs using multiple in silico metrics and characterize the strengths and limitations of its generalisation capacity. Then, we steer REXzyme towards the generation of enhanced formate dehydrogenases for their reverse natural reaction, i.e., the reduction of carbon dioxide. All 20 generated enzymes were expressed and purified as soluble proteins and exhibited CO2-reduction activity, with the most active design reaching 2.5-fold the specific activity of the reference enzyme. Notably, all substitutions in the designed variants were performed on distal positions outside the catalytic site. Our results show that reaction-to-sequence translation provides a route for efficiently engineering enhanced biocatalysts.

biochemistry↗

bioRxiv2026

ProtGPT3: an Open-source family of Promptable and Aligned Protein Language Models

Generative protein language models (pLMs) enable exploration of vast sequence spaces for protein design, but reliably controlling generation toward desired functional families remains challenging. While protein generation has broadly followed trends in NLP, two directions remain underexplored: alignment methods that optimize model behavior toward design objectives, and prompting-based control at inference time without fine-tuning. We introduce ProtGPT3, an open-source family of protein language models spanning 112M to 10B parameters and integrated with the Hugging Face ecosystem. The suite includes both single-sequence and multiple sequence alignment (MSA)-promptable models, enabling flexible conditioning for generation. Across model scales and protein families, we systematically compare supervised fine-tuning and few-shot prompting using homologous sequences. Analogous to how large language models (LLMs) are routinely aligned with user intent, we study post-training alignment in single-sequence models using sequence-complexity and structure-confidence metrics across the proteome. We find that alignment reduces low-complexity generations while preserving sequence diversity. Furthermore, we show that few-shot prompting is a competitive and more scalable alternative to supervised fine-tuning for controlled generation. In a low-data defluorinase case study, ProtGPT3-MSA achieved higher computational success rates than fine-tuned baselines and produced designs that were soluble and expressed following experimental validation. Finally, we explore the potential of inference-time compute in MSA models by introducing a homolog-based Feynman-Kac inference procedure for steering protein generation toward desired targets. All models are publicly available at https://huggingface.co/collections/AI4PD/protgpt3-family.

bioengineering↗

bioRxiv2026

Generalise or Memorise? Benchmarking Ligand-Conditioned Protein Generation from Sequence-Only Data

Proteins can bind small molecules with high specificity. However, designing proteins that bind userdefined ligands remains a challenge, typically relying on structural information and costly experimental iteration. While protein language models (pLMs) have shown promise for unconditional generation and conditioning on coarse functional labels, instance-level conditioning on a specific ligand has not been evaluated using purely textual inputs. Here we frame small-molecule protein binder design as a sequence-to-sequence translation problem and train ligand-conditioned pLMs that map molecular strings to candidate binder sequences. We curate large-scale ligand-protein datasets (>17M ligand-protein pairs) covering different data regimes and train a suite of models, spanning 16 to 700M parameters. Results reveal a consistent trade-off driven by supervision ambiguity: when each ligand is paired with few proteins, models generate near-neighbour, foldable sequences; when each ligand is paired with many proteins, generations are more diverse but less consistently foldable. Our study exposes how annotation diversity and sampling choices elicit this behaviour and how it changes with the data distribution. These insights highlight dataset redundancy and incompleteness as key bottlenecks for sequence-only binder design. We release the curated datasets, trained models, and evaluation tools to support future work on ligand-conditioned protein generation.

bioinformatics↗

bioRxiv2026

Designing AI-programmable therapeutics with the EDEN family of foundation models

The ability to interpret, modify, and design DNA has driven many of the most significant advances in modern medicine, from diagnostics, biologics, and vaccines to cell and gene therapies. However, the inherent complexity of biological systems means that most modern medicines are still engineered using bespoke, labor-intensive processes. To address the need for a generalisable and programmable approach to therapeutic design, we introduce the EDEN (environmentally-derived evolutionary network) family of metagenomic foundation models, including a 28 billion parameter model trained on 9.7 trillion nucleotide tokens from BaseData1. This dataset, at the time of training, contained more than 10 billion novel genes from over 1 million new species, and is intentionally enriched for environmental and host-associated metagenomes, phage sequences, and mobile genetic elements, enabling the model to learn from diverse and novel cross-species evolutionary mechanisms and apply them to key challenges in human health. EDEN achieves state-of-the-art performance across a series of predictive and generative genomic and protein benchmarks. To demonstrate the models broad applicability across biology, we evaluate EDENs capacity for programmable therapeutic design by challenging a single architecture to design biological novelty across three distinct therapeutic modalities, disease areas and biological scales: (i) large gene insertion, (ii) antibiotic peptide design, and (iii) microbiome design. First, we demonstrate AI-programmable Gene Insertion (aiPGI), in which EDEN designs de novo large serine recombinases (LSRs) capable of inserting large pieces of DNA at desired target sites in the human genome when prompted only on 30 nucleotides of DNA sequence from the desired target site. In low-N experimental validation, EDEN generated multiple active recombinases for all tested disease-associated genomic loci (ATM, DMD, F9, FANCC, GALC, IDS, P4HA1, PHEX, RYR2, USH2A) and 4 potential safe harbor sites in the human genome. EDEN achieves an overall functional hit rate of 63.2% across diverse DNA prompts when prompted on only 30bp of DNA from outside the training data. 50% of EDEN-generated LSRs were active in human cells, achieving therapeutically relevant levels of CAR insertion in primary human T cells. We also show that EDEN can generate active bridge recombinases when prompted on the associated guide RNA alone, with sequence identities to training and public data as low as 65%. These results pave the way for a new generation of cell and gene therapies by opening the door to rapid, programmable and site-specific integration of large genetic payloads without double-strand breaks. This offers an alternative to the safety, efficiency and payload limitations inherent in viral or nuclease-based editing at thousands of currently intractable human therapeutic targets. Second, we use the same model to generate a focused low-N library of novel antimicrobial peptides where 97% showed activity, with top candidates achieving single-digit micromolar potency against critical-priority multidrug-resistant pathogens. Third, to demonstrate that EDEN captures inter-genomic features, we design a gigabase-scale microbiome with over 94,000 synthetic metagenomic assemblies, including prophage genomes and correct cross-species metabolic pathway completions. The EDEN-generated synthetic microbiome covers 9,067 species with a biome-specific taxonomic accuracy of 99%. Over 1,500 of the generated species were outside the fine-tuning dataset while retaining the correct microecological properties and biome association, thus significantly expanding genetic and taxonomic diversity. Together, these results establish a new strategic direction for AI-programmable therapeutics, in which a single foundation model architecture designs candidate therapeutics across diverse modalities and disease areas. This suggests that the combination of billions of years of evolutionary data with specific therapeutic records offers a clear, scaling-driven path to making therapeutic design a predictable engineering discipline. O_FIG O_LINKSMALLFIG WIDTH=141 HEIGHT=200 SRC="FIGDIR/small/699009v2_ufig1.gif" ALT="Figure 1"> View larger version (59K): org.highwire.dtl.DTLVardef@68c20borg.highwire.dtl.DTLVardef@19b8e8corg.highwire.dtl.DTLVardef@1ab9362org.highwire.dtl.DTLVardef@1592cb7_HPS_FORMAT_FIGEXP M_FIG C_FIG

genomics↗

bioRxiv2025

Crowdsourced Protein Design: Lessons From the Adaptyv EGFR Binder Competition

In this report, we summarize and analyze the 2024 Adaptyv protein design competition. Participants used computational and Machine Learning (ML) methods of their choice to design proteins that bind the Epidermal Growth Factor Receptor (EGFR), a key drug target involved in cell growth, differentiation, and cancer development. Over 1,800 designs were submitted across two rounds. Of these, 601 proteins were selected and characterized for expression and binding affinity to EGFR, with competitors both optimizing existing binders (KD = 1.21 nM) and creating de novo binders (KD = 82 nM). All selected designs were experimentally validated using Adaptyvs automated Bio-Layer Interferometry (BLI) pipeline. This competition illustrates the potential of crowdsourcing to drive creativity and innovation in protein design. However, it also exposed key challenges, such as the lack of standardized benchmarks, experimental design targets, and robust computational metrics for method comparison. We anticipate that future competitions will address these gaps and further motivate progress in computational protein design.

bioengineering↗

bioRxiv2024

De novo design of triosephosphate isomerases using generative language models

The design of proteins with tailored functions is of immense interest to biotechnology, medicine, and the chemical industry. While protein design is rapidly evolving with the use of AI techniques, the design of complex enzymes remains a challenge. Here, we present the use of two large language models (LLMs), ZymCTRL and ProtGPT2, for the generation of de novo enzymes that catalyze the triosephosphate isomerase (TIM) reaction. Natural TIM enzymes are obligatory oligomers that catalyze a multi-step isomerization reaction near the diffusion limit. This makes TIM an ideal target to assess the generative ability of protein language models. Newly generated sequences were filtered to obtain a set of twelve candidates from each approach for experimental validation. Multiple constructs from both language models exhibit the intended function in vivo through their ability to complement a TIM-deficient E. coli strain. In-depth characterization of the best-behaving artificial enzyme reveals behavior and catalytic efficiency close to its natural counterparts. These findings support the use of conditional and fine-tuned unconditional LLMs for the generation of complex enzymes.

biochemistry↗

bioRxiv2024

Conditional language models enable the efficient design of proficient enzymes

The design of functional enzymes holds promise for transformative solutions across various domains but presents significant challenges. Inspired by the success of language models in generating nature-like proteins, we explored the potential of an enzyme-specific language model in designing catalytically active artificial enzymes. Here, we introduce ZymCTRL (enzyme control), a conditional language model trained on the enzyme sequence space, capable of generating enzymes based on user-defined specifications. Experimental validation at diverse data regimes and for different enzyme families demonstrated ZymCTRLs ability to generate active enzymes across various sequence identity ranges. Specifically, we describe the design of carbonic anhydrases and lactate dehydrogenases in zero-shot, without requiring further training of the model, and showcasing activity at sequence identities below 40% compared to natural proteins. Biophysical analysis confirmed the globularity and well-folded nature of the generated sequences. Furthermore, fine-tuning the model enabled the generation of lactate dehydrogenases outside of natural sequence space but with activity comparable to their natural counterparts. Two of the artificial lactate dehydrogenases were selected for scale production and successfully lyophilised, maintaining activity and demonstrating preliminary conversion in one-pot enzymatic cascades under extreme conditions. Our findings open a new door towards the rapid and cost-effective design of artificial proficient enzymes. The model and dataset are freely available to the community.

bioinformatics↗

bioRxiv2024

Code for Collagen Folding Deciphered

Collagen triple helix folds in two steps: nucleation of three polypeptides at the C-termini followed by zip-chain like propagation. The triple helices found in all domains of life as well as viruses contain upto 6000 amino acids in each polypeptide that are also frequently interrupted with non-helical sequences that disrupt folding and reduce stability. Given the length of polypeptide and the disruptive interruptions, compensating mechanisms that stabilize against local unfolding during propagation and offset the entropic cost of folding the long polypeptides are not fully understood. Here, we show that the information for correct folding of collagen triple helices is encoded in their sequence as interchain electrostatic interactions. In case of humans, disrupting these interactions causes severe to lethal diseases. Key ResultCollagen triple helices found in all the three domains of life as well as viruses have converged on similar mechanism to fold correctly.

biochemistry↗

bioRxiv2022

From sequence to function through structure: deep learning for protein design

The process of designing biomolecules, in particular proteins, is witnessing a rapid change in available tooling and approaches, moving from design through physicochemical force fields, to producing plausible, complex sequences fast via end-to-end differentiable statistical models. To achieve conditional and controllable protein design, researchers at the interface of artificial intelligence and biology leverage advances in natural language processing (NLP) and computer vision techniques, coupled with advances in computing hardware to learn patterns from growing biological databases, curated annotations thereof, or both. Once learned, these patterns can be leveraged to provide novel insights into mechanistic biology and the design of biomolecules. However, navigating and understanding the practical applications for the many recent protein design tools is complex. To facilitate this, we 1) document recent advances in deep learning (DL) assisted protein design from the last three years, 2) present a practical pipeline that allows to go from de novo-generated sequences to their predicted properties and web-powered visualization within minutes, and 3) leverage it to suggest a generated protein sequence which might be used to engineer a biosynthetic gene cluster to produce a molecular glue-like compound. Lastly, we discuss challenges and highlight opportunities for the protein design field. AvailabilitypLM generated and UniRef50 sampled sequence sets and predictions are available at http://data.bioembeddings.com/public/design. Code-base and Notebooks for analysis are available at https://github.com/hefeda/PGP. An online version of Table 1 can be found at https://github.com/hefeda/design_tools. O_TBL View this table: org.highwire.dtl.DTLVardef@1823c4borg.highwire.dtl.DTLVardef@1449504org.highwire.dtl.DTLVardef@193458borg.highwire.dtl.DTLVardef@1bab4c4org.highwire.dtl.DTLVardef@b1d39f_HPS_FORMAT_FIGEXP M_TBL O_FLOATNOTable 1:C_FLOATNO O_TABLECAPTIONmachine-learning-based protein design methods. Methods ordered by their release date, accounting for the date of pre-prints when available. Class (1-4) captured in Fig. 1. Expanded method detail in main text. Unnamed methods are referenced by the first name of the first author. Online version at https://github.com/hefeda/design_tools. Legend: Architecture: The architecture of the deep learning model; E2E: an end-to-end differentiable solution; Input: the input to run (infer from) the model, e.g. a contact map; Output: the output, e.g. a protein sequence; Dataset: the number and type of samples used to train the method; Params: the exact or estimated number of parameters of the model; 1: the 3D recovery was performed in an external, second step; 2: conditioned generation; -: no input required for generation. C_TABLECAPTION C_TBL

bioinformatics↗

bioRxiv2022

A deep unsupervised language model for protein design

Protein design aims to build new proteins from scratch thereby holding the potential to tackle many environmental and biomedical problems. Recent progress in the field of natural language processing (NLP) has enabled the implementation of ever-growing language models capable of understanding and generating text with human-like capabilities. Given the many similarities between human languages and protein sequences, the use of NLP models offers itself for predictive tasks in protein research. Motivated by the evident success of generative Transformer-based language models such as the GPT-x series, we developed ProtGPT2, a language model trained on protein space that generates de novo protein sequences that follow the principles of natural ones. In particular, the generated proteins display amino acid propensities which resemble natural proteins. Disorder and secondary structure prediction indicate that 88% of ProtGPT2-generated proteins are globular, in line with natural sequences. Sensitive sequence searches in protein databases show that ProtGPT2 sequences are distantly related to natural ones, and similarity networks further demonstrate that ProtGPT2 is sampling unexplored regions of protein space. AlphaFold prediction of ProtGPT2-sequences yielded well-folded non-idealized structures with embodiments as well as large loops and revealed new topologies not captured in current structure databases. ProtGPT2 has learned to speak the protein language. It has the potential to generate de novo proteins in a high throughput fashion in a matter of seconds. The model is easy-to-use and freely available.

bioinformatics↗

bioRxiv2020

Protlego: A Python package for the analysis and design of chimeric proteins

MotivationGene duplication and recombination of protein fragments have led to the highly diverse protein space that we observe today. By mimicking this natural process, the design of protein chimeras via fragment recombination has proven experimentally successful and has opened a new era for the design of customizable proteins. The in-silico building of structural models for these chimeric proteins, however, remains a manual task that requires a considerable degree of expertise and is not amenable for high-throughput studies. Energetic and structural analysis of the designed proteins often require the use of several tools, each with their unique technical difficulties and available in different programming languages or web servers. ResultsWe have implemented a Python package that enables automated, high-throughput design of chimeras and their structural analysis. First, it is possible to fetch evolutionarily conserved fragments from a built-in database (also available at fuzzle.uni-bayreuth.de). These relationships can then be represented via networks or further selected for chimera construction via recombination. Designed chimeras or natural proteins are then scored and minimised with the Charmm and Amber forcefields and their diverse structural features can be analysed at ease. Here, we showcase Protlegos pipeline by exploring the relationships between the P-loop and Rossmann superfolds and building and characterising their offspring chimeras. We believe that Protlego provides a powerful new tool for the protein design community. Availability and implementationProtlego is freely available at (https://hoecker-lab.github.io/protlego/) with tutorials and documentation.

bioinformatics↗