arXiv subjects

Frances H. Arnold

Publications and source records attributed to Frances H. Arnold.

12 recordsLinked to original sources

arXiv2026

OXtal: An All-Atom Diffusion Model for Organic Crystal Structure Prediction

Accurately predicting experimentally realizable 3D molecular crystal structures from their 2D chemical graphs is a long-standing open challenge in computational chemistry called crystal structure prediction (CSP). Efficiently solving this problem has implications ranging from pharmaceuticals to organic semiconductors, as crystal packing directly governs the physical and chemical properties of organic solids. In this paper, we introduce OXtal, a large-scale 100M parameter all-atom diffusion model that directly learns the conditional joint distribution over intramolecular conformations and periodic packing. To efficiently scale OXtal, we abandon explicit equivariant architectures imposing inductive bias arising from crystal symmetries in favor of data augmentation strategies. We further propose a novel crystallization-inspired lattice-free training scheme, Stoichiometric Stochastic Shell Sampling ($S^4$), that efficiently captures long-range interactions while sidestepping explicit lattice parametrization -- thus enabling more scalable architectural choices at all-atom resolution. By leveraging a large dataset of 600K experimentally validated crystal structures (including rigid and flexible molecules, co-crystals, and solvates), OXtal achieves orders-of-magnitude improvements over prior ab initio machine learning CSP methods, while remaining orders of magnitude cheaper than traditional quantum-chemical approaches. Specifically, OXtal recovers experimental structures with conformer $\text{RMSD}_1<0.5$ Å and attains over 80\% packing similarity rate, demonstrating its ability to model both thermodynamic and kinetic regularities of molecular crystallization.

cs.LG↗

arXiv2026

General Multimodal Protein Design Enables DNA-Encoding of Chemistry

Evolution is an extraordinary engine for enzymatic diversity, yet the chemistry it has explored remains a narrow slice of what DNA can encode. Deep generative models can design new proteins that bind ligands, but none have created enzymes without pre-specifying catalytic residues. We introduce DISCO (DIffusion for Sequence-structure CO-design), a multimodal model that co-designs protein sequence and 3D structure around arbitrary biomolecules, as well as inference-time scaling methods that optimize objectives across both modalities. Conditioned solely on reactive intermediates, DISCO designs diverse heme enzymes with novel active-site geometries. These enzymes catalyze new-to-nature carbene-transfer reactions, including alkene cyclopropanation, spirocyclopropanation, B-H, and C(sp$^3$)-H insertions, with high activities exceeding those of engineered enzymes. Random mutagenesis of a selected design further confirmed that enzyme activity can be improved through directed evolution. By providing a scalable route to evolvable enzymes, DISCO broadens the potential scope of genetically encodable transformations. Code is available at https://github.com/DISCO-design/DISCO.

cs.LG↗

arXiv2025

Steering Generative Models with Experimental Data for Protein Fitness Optimization

Protein fitness optimization involves finding a protein sequence that maximizes desired quantitative properties in a combinatorially large design space of possible sequences. Recent advances in steering protein generative models (e.g., diffusion models and language models) with labeled data offer a promising approach. However, most previous studies have optimized surrogate rewards and/or utilized large amounts of labeled data for steering, making it unclear how well existing methods perform and compare to each other in real-world optimization campaigns where fitness is measured through low-throughput wet-lab assays. In this study, we explore fitness optimization using small amounts (hundreds) of labeled sequence-fitness pairs and comprehensively evaluate strategies such as classifier guidance and posterior sampling for guiding generation from different discrete diffusion models of protein sequences. We also demonstrate how guidance can be integrated into adaptive sequence selection akin to Thompson sampling in Bayesian optimization, showing that plug-and-play guidance strategies offer advantages over alternatives such as reinforcement learning with protein language models. Overall, we provide practical insights into how to effectively steer modern generative models for next-generation protein fitness optimization.

q-bio.BM↗

arXiv2025

CARE: a Benchmark Suite for the Classification and Retrieval of Enzymes

Enzymes are important proteins that catalyze chemical reactions. In recent years, machine learning methods have emerged to predict enzyme function from sequence; however, there are no standardized benchmarks to evaluate these methods. We introduce CARE, a benchmark and dataset suite for the Classification And Retrieval of Enzymes (CARE). CARE centers on two tasks: (1) classification of a protein sequence by its enzyme commission (EC) number and (2) retrieval of an EC number given a chemical reaction. For each task, we design train-test splits to evaluate different kinds of out-of-distribution generalization that are relevant to real use cases. For the classification task, we provide baselines for state-of-the-art methods. Because the retrieval task has not been previously formalized, we propose a method called Contrastive Reaction-EnzymE Pretraining (CREEP) as one of the first baselines for this task and compare it to the recent method, CLIPZyme. CARE is available at https://github.com/jsunn-y/CARE/.

q-bio.BM↗

arXiv2021

Protein sequence design with deep generative models

Protein engineering seeks to identify protein sequences with optimized properties. When guided by machine learning, protein sequence generation methods can draw on prior knowledge and experimental efforts to improve this process. In this review, we highlight recent applications of machine learning to generate protein sequences, focusing on the emerging field of deep generative methods.

q-bio.QM↗

arXiv2020

Machine learning-assisted directed protein evolution with combinatorial libraries

To reduce experimental effort associated with directed protein evolution and to explore the sequence space encoded by mutating multiple positions simultaneously, we incorporate machine learning in the directed evolution workflow. Combinatorial sequence space can be quite expensive to sample experimentally, but machine learning models trained on tested variants provide a fast method for testing sequence space computationally. We validate this approach on a large published empirical fitness landscape for human GB1 binding protein, demonstrating that machine learning-guided directed evolution finds variants with higher fitness than those found by other directed evolution approaches. We then provide an example application in evolving an enzyme to produce each of the two possible product enantiomers (stereodivergence) of a new-to-nature carbene Si-H insertion reaction. The approach predicted libraries enriched in functional enzymes and fixed seven mutations in two rounds of evolution to identify variants for selective catalysis with 93% and 79% ee. By greatly increasing throughput with in silico modeling, machine learning enhances the quality and diversity of sequence solutions for a protein engineering problem.

q-bio.BM↗

arXiv2019

Machine learning-guided directed evolution for protein engineering

Machine learning (ML)-guided directed evolution is a new paradigm for biological design that enables optimization of complex functions. ML methods use data to predict how sequence maps to function without requiring a detailed model of the underlying physics or biological pathways. To demonstrate ML-guided directed evolution, we introduce the steps required to build ML sequence-function models and use them to guide engineering, making recommendations at each stage. This review covers basic concepts relevant to using ML for protein engineering as well as the current literature and applications of this new engineering paradigm. ML methods accelerate directed evolution by learning from information contained in all measured variants and using that information to select sequences that are likely to be improved. We then provide two case studies that demonstrate the ML-guided directed evolution process. We also look to future opportunities where ML will enable discovery of new protein functions and uncover the relationship between protein sequence and function.

q-bio.BM↗

arXiv2007

Evolution favors protein mutational robustness in sufficiently large populations

BACKGROUND: An important question is whether evolution favors properties such as mutational robustness or evolvability that do not directly benefit any individual, but can influence the course of future evolution. Functionally similar proteins can differ substantially in their robustness to mutations and capacity to evolve new functions, but it has remained unclear whether any of these differences might be due to evolutionary selection for these properties. RESULTS: Here we use laboratory experiments to demonstrate that evolution favors protein mutational robustness if the evolving population is sufficiently large. We neutrally evolve cytochrome P450 proteins under identical selection pressures and mutation rates in populations of different sizes, and show that proteins from the larger and thus more polymorphic population tend towards higher mutational robustness. Proteins from the larger population also evolve greater stability, a biophysical property that is known to enhance both mutational robustness and evolvability. The excess mutational robustness and stability is well described by existing mathematical theories, and can be quantitatively related to the way that the proteins occupy their neutral network. CONCLUSIONS: Our work is the first experimental demonstration of the general tendency of evolution to favor mutational robustness and protein stability in highly polymorphic populations. We suggest that this phenomenon may contribute to the mutational robustness and evolvability of viruses and bacteria that exist in large populations.

q-bio.PE↗

arXiv2006

Inferring interactions from combinatorial protein libraries

Proteins created by combinatorial methods in vitro are an important source of information for understanding sequence-structure-function relationships. Alignments of folded proteins from combinatorial libraries can be analyzed using methods developed for naturally occurring proteins, but this neglects the information contained in the unfolded sequences of the library. We introduce two algorithms, logistic regression and excess information analysis, that use both the folded and unfolded sequences and compare them against contingency table and statistical coupling analysis, which only use the former. The test set for this benchmark study is a library of fictitious proteins that fold according to a hypothetical energy model. Of the four methods studied, only logistic regression is able to correctly recapitulate the energy model from the sequence alignment. The other algorithms predict spurious interactions between alignment positions with strong but individual influences on protein stability. When present in the same protein, stabilizing amino acids tend to lower the energy below the threshold needed for folding. As a result, their frequencies in the alignment can be correlated even if the positions do not interact. We believe any algorithm that neglects the nonlinear relationship between folding and energy is susceptible to this error.

q-bio.BM↗

arXiv2005

Why highly expressed proteins evolve slowly

Much recent work has explored molecular and population-genetic constraints on the rate of protein sequence evolution. The best predictor of evolutionary rate is expression level, for reasons which have remained unexplained. Here, we hypothesize that selection to reduce the burden of protein misfolding will favor protein sequences with increased robustness to translational missense errors. Pressure for translational robustness increases with expression level and constrains sequence evolution. Using several sequenced yeast genomes, global expression and protein abundance data, and sets of paralogs traceable to an ancient whole-genome duplication in yeast, we rule out several confounding effects and show that expression level explains roughly half the variation in Saccharomyces cerevisiae protein evolutionary rates. We examine causes for expression's dominant role and find that genome-wide tests favor the translational robustness explanation over existing hypotheses that invoke constraints on function or translational efficiency. Our results suggest that proteins evolve at rates largely unrelated to their functions, and can explain why highly expressed proteins evolve slowly across the tree of life.

q-bio.PE↗

arXiv2005

Why high-error-rate random mutagenesis libraries are enriched in functional and improved proteins

Recently, several groups have used error-prone polymerase chain reactions to construct mutant libraries containing up to 27 nucleotide mutations per gene on average, and reported a striking observation: although retention of protein function initially declines exponentially with mutations as has previously been observed, orders of magnitude more proteins remain viable at the highest mutation rates than this trend would predict. Mutant proteins having improved or novel activity were isolated disproportionately from these heavily mutated libraries, leading to the suggestion that distant regions of sequence space are enriched in useful cooperative mutations and that optimal mutagenesis should target these regions. If true, these claims have profound implications for laboratory evolution and for evolutionary theory. Here, we demonstrate that properties of the polymerase chain reaction can explain these results and, consequently, that average protein viability indeed decreases exponentially with mutational distance at all error rates. We show that high-error-rate mutagenesis may be useful in certain cases, though for very different reasons than originally proposed, and that optimal mutation rates are inherently protocol-dependent. Our results allow optimal mutation rates to be found given mutagenesis conditions and a protein of known mutational robustness.

q-bio.QM↗

arXiv2004

Thermodynamic Prediction of Protein Neutrality

We present a simple theory that uses thermodynamic parameters to predict the probability that a protein retains the wildtype structure after one or more random amino acid substitutions. Our theory predicts that for large numbers of substitutions the probability that a protein retains its structure will decline exponentially with the number of substitutions, with the severity of this decline determined by properties of the structure. Our theory also predicts that a protein can gain extra robustness to the first few substitutions by increasing its thermodynamic stability. We validate our theory with simulations on lattice protein models and by showing that it quantitatively predicts previously published experimental measurements on subtilisin and our own measurements on variants of TEM1 beta-lactamase. Our work unifies observations about the clustering of functional proteins in sequence space, and provides a basis for interpreting the response of proteins to substitutions in protein engineering applications.

q-bio.BM↗