AlphaProtein Novo: generative diffusion pipeline for de novo enzyme design

Google DeepMind released AlphaProtein Novo (AP Novo), a generative diffusion pipeline for de novo enzyme design via structural motif scaffolding that co-generates protein structures and sequences conditioned on a catalytic motif and ligand context. The package combines the diffusion model with optional LigandMPNN sequence redesign, AlphaFold 3 structure prediction, and evaluation metrics, orchestrated by run_pipeline.py, and requires pretrained generator weights (generator.bin.zst) downloaded separately from Google Cloud Storage under a separate terms-of-use agreement. Findings from the code, model parameters, or outputs must cite the bioRxiv paper "Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo.

AlphaProtein Novo AP Novo is a generative diffusion pipeline for de novo enzyme design via structural motif scaffolding. This package includes a diffusion model which co-generates protein structures and sequences conditioned on a catalytic motif and ligand context. This is run using run generator.py . To form an end-to-end pipeline, the diffusion model is combined with optional sequence redesign using LigandMPNN run ligandmpnn.py , structure prediction using AlphaFold 3 run alphafold.py , and evaluation metrics evaluate design.py . The run pipeline.py script runs all stages in sequence. See the Quickstart quickstart section for a basic launch command and Example Design Campaigns example-design-campaigns for more detailed examples. Any publication that discloses findings arising from using this source code, the model parameters, or outputs produced by those should cite citing-this-work the Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1 paper. 📒 Note: Pretrained model weights are not part of this package and must be downloaded from Google Cloud Storage see Model Parameters model-parameters-weights below . Use is subject to these terms of use https://github.com/google-deepmind/alphaprotein-novo/blob/main/WEIGHTS TERMS OF USE.md . Scripts look for weights under ./models/apnovo generator by default, or you can point to another directory using --apn model dir for run pipeline.py , --model dir for run generator.py , or settings.model dir in your design manifest. Two Python environments are recommended: - Primary Environment alphaprotein novo : Contains JAX. Runs diffusion generation, AlphaFold 3 folding, evaluation metrics, and the pipeline orchestrator. - LigandMPNN Environment ligandmpnn Optional : Contains PyTorch. Used to run LigandMPNN for sequence design. Create and activate a Python 3.12 environment uv venv --python 3.12 .venv source .venv/bin/activate Install JAX with CUDA 12 support or CPU: uv pip install -U jax uv pip install -U "jax cuda12 =0.4.30" Install AlphaFold 3 uv pip install git+https://github.com/google-deepmind/alphafold3.git Install AP Novo in editable mode cd /path/to/alphaprotein novo uv pip install -e . Compile AlphaFold 3 chemical component data CCD pickle build data 📒 Tip: build data compiles ccd.pickle and chemical component sets.pickle required by AlphaFold 3 to parse ligands. If components.cif cannot be located, set the LIBCIFPP DATA DIR environment variable to its directory before running build data . Only required if you plan to run optional sequence redesign with LigandMPNN: 1. Create and activate Python 3.11 environment uv venv --python 3.11 .venv-ligandmpnn source .venv-ligandmpnn/bin/activate 2. Clone LigandMPNN and download weights git clone https://github.com/dauparas/LigandMPNN.git cd LigandMPNN bash get model params.sh ./model params 3. Install dependencies uv pip install -r requirements.txt uv pip install "setuptools<82" ProDy requires pkg resources removed in setuptools 82+ 4. Optional Set environment variables. This allows you to avoid having to provide the --ligandmpnn dir and --ligandmpnn python flags to run pipeline.py. export LIGANDMPNN DIR=/home/<your username /projects/LigandMPNN For example. export LIGANDMPNN PYTHON=/home/<your username /miniconda3/envs/ligandmpnn/bin/python For example. - AP Novo : Download pretrained generator weights generator.bin.zst into ./models/apnovo generator subject to the AP Novo Parameters Terms of Use https://github.com/google-deepmind/alphaprotein-novo/blob/main/WEIGHTS TERMS OF USE.md ; or configure a directory via --apn model dir , --model dir , or settings.model dir in your manifest : mkdir -p models/apnovo generator wget -P models/apnovo generator https://storage.googleapis.com/alphaprotein novo/generator.bin.zst - AlphaFold 3 Leaving Atom AF3-LA : Download neural network weights af3 leaving atom.bin.zst into ./models/af3 la subject to the AlphaFold 3 Parameters Terms of Use https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS TERMS OF USE.md . Structure prediction uses these by default, because AF3-LA is fine-tuned for leaving atom handling and therefore models the covalent intermediates common in AP Novo designs: mkdir -p models/af3 la wget -P models/af3 la https://storage.googleapis.com/alphafold3/af3 leaving atom.bin.zst Alternatively, stock AlphaFold 3 weights af3.bin.zst can be used, subject to the same terms of use. Pass --model dir=./models/af3 or --af3 model dir=./models/af3 to run pipeline.py to select them: mkdir -p models/af3 wget -P models/af3 https://storage.googleapis.com/alphafold3/af3.bin.zst The standard way to run AP Novo is via run pipeline.py . It executes design generation, sequence design, folding, and metrics calculation end-to-end from a single "manifest" file containing configuration settings: Ensure the primary environment is active source .venv/bin/activate python run pipeline.py \ --manifest=examples/kemp eliminase/kemp manifest.json \ --output dir=./kemp campaign \ --af3 model dir=./models/af3 la \ --ligandmpnn dir=/path/to/LigandMPNN \ --ligandmpnn python="/path/to/ligandmpnn/.venv-ligandmpnn/bin/python" 📒 Tip: To quickly verify your installation without waiting for a full-length diffusion trajectory, swap in examples/kemp eliminase/kemp test manifest.json , which runs the same Kemp eliminase benchmark with reduced sampling steps. Outputs are organized into stage-specific subdirectories under --output dir see Generated Artifacts & Reports generated-artifacts--reports below . By default, re-running the pipeline resumes interrupted campaigns by skipping completed designs; you can also force a full rerun or execute individual stages see Modular Stage Execution modular-stage-execution . Campaigns are configured using a JSON manifest describing the design problem, motif conditioning, and downstream processing. Here is an example based on examples/kemp eliminase/kemp manifest.json : { "settings": { "model dir": "./models/apnovo generator" }, "defaults": { "input file": "kemp eliminase motif.cif", "is author naming": true, "num sampling steps": 1000 }, "designs": { "name": "kemp tight", "motif str": "A1,A2,A3|10-40,{},2-30,{},2-30,{},10-40/B1", "motif atoms": "A1:NE2,ND1 A2:OD1 A3:ND2", "num designs": 200 } , "resequence": { "enabled": true, "temperature": 0.1 }, "folding": { "inputs": "resequenced" , "seeds": 230 , "states": {"name": "monomer", "ligands": }, {"name": "complex", "ligands": {"id": "B", "ccd code": "6NT"} } }, "evaluation": { "suite": "kemp eliminase" } } See below example-design-campaigns for further, full-fledged examples. - settings : Run-level configuration e.g. model dir , output dir . Overridden by corresponding CLI flags. - defaults : Default fields applied to any design job that does not explicitly set them e.g. num sampling steps , seed start . - designs : List of design tasks. Each specifies the number of designs and the input motif see Design Inputs design-inputs below . - resequence : Configures LigandMPNN sequence redesign enabled , temperature . Set "enabled": false to fold the generated sequence directly without LigandMPNN. - folding : Specifies which structures to fold inputs: "generated" or "resequenced" , RNG seeds , and target states e.g. apo monomer, ligand complex . - evaluation : Sets evaluation parameters, including the enzyme evaluation suite "kemp eliminase" , "serine esterase" , "dehp esterase" , "carbene transfer" , or "nitrene transfer" and an optional reference cif override by default, each design is scored against its own per-sample ground-truth motif structure . 📒 Note: Relative file paths in the manifest resolve relative to the directory containing the manifest file. De novo enzyme design starts with a 3D arrangement of sidechain and ligand atoms that represents a transition-state or intermediate of the target reaction. The goal of design is to generate a protein that holds this motif in the intended conformation. At a high level, you will need to decide the following: - Input motif: what are the catalytic residues and small molecule ligands that represent your reaction? What conformation should the enzyme bind them in? - Full or partial residue motif: Do you want to constrain the backbone positions of all residues in the motif, or just functional groups on the sidechains? - Indexed or unindexed motif residues: Do you want to specify the residue indexes primary sequence positions of motif residues on the design ahead of generation indexed , or have the model decide where they go unindexed ? Unindexed conditioning can in theory find more optimal motif placements, but we find that indexed conditioning gives slightly better results in practice. - Novel scaffold or partial diffusion: Do you want to generate a protein from scratch, or re-use an existing protein structure as a starting point partial diffusion ? The design problem is specified in full using the following fields in the designs section of the manifest: - input file is a path to a CIF file containing the input motif. - motif str is a string describing which parts of the input structure make up the motif and how they are arranged in the output design. - The string specifies a series of design chains delimited by / . - The first design chain contains a series of comma-separated segments representing scaffolded motif or designable regions. Subsequent chains should consist of a single residue, representing a fixed ligand. - For example, A1,5-10,A5-6,3/B1 means "generate a design that scaffolds residue 1 of chain A of the input structure, followed by 5 to 10 residues of generated protein, followed by scaffolding residues 5-6 of input chain A, followed by 3 residues of generated protein, with a 2nd chain consisting exactly of the atoms in input chain B a ligand ". - In this example, residues 1, 5, and 6 of input chain A and 1 of chain B are "fixed" to their input coordinates, and the model will try to generate coordinates for the remaining residues to best preserve the position of the fixed residues or ligands. - Designable length ranges are sampled before generation, e.g. for 5-10 , a number between 5 and 10 inclusive is chosen uniformly at random and fixed for the duration of the diffusion process. - The ordering of the motif segments in motif str can be sampled by writing "A1,A5-6|{},5-10,{},3/B1" . This will resolve to either A1,5-10,A5-6,3/B1 or A5-6,5-10,A1,3/B1 with equal probability. - The string specifies a series of design chains delimited by - seq length is an optional string constraining the total residue length of the designed protein chain either an exact integer like "150" or an inclusive range like "120-160" . When specified, designable segment lengths in motif str are sampled conditioned on the total chain length falling within this range. - unindexed motif residues is an optional string containing motif residues for "unindexed" conditioning. By default, the motif residues are at fixed, pre-defined positions along the primary sequence "indexed" conditioning . Under unindexed conditioning, the model decides during generation where the motif should go. - This is a comma-separated list of residues or residue ranges, e.g. "A1,A2,A3" or "A120-121,A188" . - If specified, motif str must contain a single designable element e.g. 10-250/B1 , with no fixed residues specified other than ligand chains which are always fixed . - This is a comma-separated list of residues or residue ranges, e.g. - motif atoms is an optional string describing which atoms of motif residues are considered fixed, allowing for side chains to be generated by the model and thereby be "flexible" . This is sometimes called "tip atom" or "atomic motif" conditioning. - This is a space-separated list of groups, where each group consists of a residue name and a comma-separated list of atom names. - If not specified, all atoms of the residues in motif str are fixed. - If specified, the indicated atoms are fixed and the rest of the residue is generated by the model. - For example, "A1:NE2,ND1 A2:OD1" means "fix NE2 and ND1 of residue A1 and OD1 of residue A2". - reseq residues is an optional string specifying residues on the motif to "resequence", or allow to change in amino-acid identity during the generation and resequence stages. This can be used to constrain the backbone position of a residue while allowing its sidechain to vary in identity. - is author naming is a boolean indicating whether the motif string uses author-assigned residue numbers or PDB "internal" numbering. PyMOL and the literature almost always use author naming. The PDB structure viewer uses internal numbering. - partial diffusion input file is an optional path to a CIF file containing a starting protein structure such as a parent design to diversify via partial diffusion. Must be specified together with partial diffusion num steps . - partial diffusion num steps is an optional integer specifying how many reverse diffusion steps out of num sampling steps to unroll when running partial diffusion from partial diffusion input file . Setting this to 600 will lead to outputs with roughly TM-score 95 to the parent design; 900 steps will lead to TM-score ~80. Total number of denoising steps is 1000. The examples/ directory contains manifest files and input motif structures .cif that partially reproduce the settings used for the best designs in the AlphaProtein Novo paper https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1 . Problem-specific evaluation metrics are also included in the code for each of these examples, and are invoked by the manifests using the evaluation.suite field. Note that this repo is a port of the original Google-internal pipeline used to generate the designs in the paper. Some settings e.g. numbers of resequences and folding seeds have been reduced relative to the paper so that example runs can complete in a reasonable time. Run any of the example campaigns end-to-end using, for example: python run pipeline.py \ --manifest=examples/kemp eliminase/kemp manifest.json \ --output dir=/tmp/apn example kemp This assumes that AP Novo and AF3 weights are in the default locations and LigandMPNN environment variables are set. Example manifests: 1. Kemp eliminase examples/kemp eliminase/ - Motif contains glutamate catalytic base and serine oxyanion hole H-bond donor with 6-nitrobenzotriazole transition-state analog. Used for the design of GDM KE 1483 . - Main pipeline: kemp manifest.json . Unindexed conditioning variant: kemp unindexed manifest.json . Fast testing variant with 50 denoising steps will not produce good designs : kemp test manifest.json . - Folds monomer apo and 6NT-bound states with 5 AF3 seeds each. - Computes self-consistency, pocket RMSD, and Kemp eliminase catalytic geometry metrics. 2. Motif contains glutamate catalytic base and serine oxyanion hole H-bond donor with 6-nitrobenzotriazole transition-state analog. Used for the design of 3. 4MU-Ac serine esterase examples/serine esterase/ - Motif is derived from cutinase 1xzm, with a Ser-His-Asp catalytic triad, three oxyanion hole H-bond donors, and 4-methylumbelliferyl acetate tetrahedral intermediate. Used for the design of GDM SE 2937 . - AF3 folding across all five reaction states monomer , es , complex / TI1, aei , and ti2 , with evaluation of catalytic triad/oxyanion hole H-bonds and cross-state ES/TI1 ligand RMSD and coordinate standard deviation. 4. Motif is derived from cutinase 1xzm, with a Ser-His-Asp catalytic triad, three oxyanion hole H-bond donors, and 4-methylumbelliferyl acetate tetrahedral intermediate. Used for the design of 5. DEHPase — novel scaffold examples/dehp esterase denovo/ - Motif is derived from kexin 1r64, with a Ser-His-Asp catalytic triad, 2 oxyanion hole H-bond donors, and bis 2-ethylhexyl phthalate tetrahedral intermediate, used to design GDM DEHP 0176 . - AF3 folding in 5 reaction states as in 4MU-Ac esterase example. 6. Motif is derived from kexin 1r64, with a Ser-His-Asp catalytic triad, 2 oxyanion hole H-bond donors, and bis 2-ethylhexyl phthalate tetrahedral intermediate, used to design 7. DEHPase — partial diffusion examples/dehp esterase partial diffusion/ - Partial-diffusion backtracking 575 steps, out of 1000 from parent design GDM DEHP 0176.cif , used to design GDM DEHP 0376 . - Same folding and evaluation setup as in 4MU-Ac and novel-scaffold DEHPase examples above. 8. Partial-diffusion backtracking 575 steps, out of 1000 from parent design 9. Carbene transferase examples/carbene transferase/ - Motif with axial histidine, heme cofactor HEM , and 1S,2S -cyclopropanation transition-state/product conformer, used to create GDM CT 0103 . - Folding in 2 states: monomer apo and complex HEM + product . Evaluation computes self-consistency, pocket-aligned ligand RMSD, and carbene transferase heme-pocket geometry metrics. 10. Motif with axial histidine, heme cofactor 11. Piperidine synthase / nitrene transferase examples/nitrene transferase/ - Motif with axial histidine and heme-substrate cofactor complex, used to create GDM NT 0151 . - Folding in 7 states: monomer apo , complex , heme substrate , heme piperidine R , heme pyrrolidine S , fiveazidopentylbenzene heme 1 , fiveazidopentylbenzene heme 2 . Evaluation computes cross-state Fe/N self-RMSD, Fe–N distances, and regioselectivity based on near-attack atom distances piperidine propensity . 12. Motif with axial histidine and heme-substrate cofactor complex, used to create 📒 Note: use side chain context : By default, run ligandmpnn.py and resequence.use side chain context enable fixed-residue side-chain atom context --ligand mpnn use side chain context=1 , which improves design outcomes. In these example manifests, "use side chain context": false is set to reflect the exact settings used in the paper, but for your own designs you should generally enable side-chain context. run pipeline.py organizes outputs by pipeline stage: kemp campaign/ ├── pipeline index.json Campaign manifest snapshot, hash, and run metadata ├── logs/ Per-stage stdout and stderr logs │ ├── generation.log │ ├── resequence.log │ ├── folding.log │ └── evaluation.log ├── metadata/ Run-level per-design metadata <prefix metadata.json and designs.json ├── 01 generation/ Stage 1: Generated structures .cif/.pdb , motifs motif.cif , and sequences .fa ├── 02 resequence/ Stage 2: Resequenced structures reseq.cif and sequences .fa ├── 03 folded/ Stage 3: AlphaFold 3 predicted models, inputs, and confidence JSONs └── 04 eval/ Stage 4: Per-design evaluation JSONs and evaluation summary.csv All files generated in Stage 1 share the prefix <prefix <job <design num , e.g. kemp 0000 . When sequence redesign is enabled, Stage 2 appends seq<NN e.g. kemp 0000 seq00 , which becomes the design prefix used in Stages 3 and 4. | Stage | Artifact | Description | |---|---|---| | Metadata | metadata/<prefix metadata.json | Run-level design metadata fixed residues, seeds, motif paths, and folding states . | | Generation | <prefix .cif , <prefix .pdb | Generated backbone structure with ligand coordinates. | | | <prefix motif.cif | Ground-truth motif structure re-indexed to match the generated design. | | | <prefix .fa | Single-letter de novo amino acid sequence. | | Resequence | <prefix seq<NN reseq.cif | Scaffold structure with LigandMPNN-redesigned sequence. | | | <prefix seq<NN .fa , seqs/<prefix .fa | Individual and combined LigandMPNN-redesigned sequences. | | Folding | <prefix <state folded seed<N .cif | AlphaFold 3 predicted structure for the specified state and seed index <N . | | | <prefix <state confidences seed<N .json | AlphaFold 3 confidence scores plddt , ptm , iptm , ranking confidence , per atom plddt , chain pair pde mean and the seed used for folding. | | | <prefix <state af3 input.json | AlphaFold 3 folding input JSON for the specified state across all configured folding seeds. | | Evaluation | <prefix evaluation.json | Flat dictionary of metrics metrics for the individual design across folded states and seeds. | - 04 eval/evaluation summary.csv : Tabular spreadsheet written by Stage 4 with one row per design containing all scalar aggregated and per-seed metrics across states. - metadata/designs.json : Index written by Stage 1 mapping generated designs to seeds, prefixes, and completion status. - pipeline index.json : Campaign-level record written by run pipeline.py with the manifest path, SHA-256 hash, completed stages, and design prefixes. run pipeline.py tracks completed work and supports partial or stage-by-stage execution, and each stage script can also be invoked directly. - Resuming interrupted runs : By default --resume=true , re-running the pipeline command skips designs that have already finished in --output dir . - Forced rerun : Pass --noresume to rerun all stages from scratch. - Start from specific stage : Pass --from stage to resume execution starting from a specific stage generation , resequence , folding , evaluation . - Run single stage : Pass --only stage=folding to run only that stage against existing outputs on disk. The pipeline automatically scales across available hardware on a single node or across manually sharded workers: - Multi-GPU Sharding Stages 1 & 3 : When multiple GPUs are visible via CUDA VISIBLE DEVICES or detected via nvidia-smi , run pipeline.py —as well as standalone invocations of run generator.py and run alphafold.py —automatically spawns one worker process per GPU --worker id=w --num workers=N with CUDA VISIBLE DEVICES pinned to each device and distributes pending designs across workers. To restrict which GPUs are used, set CUDA VISIBLE DEVICES e.g. CUDA VISIBLE DEVICES=0,1 . - Multi-CPU Parallelism Stages 2 & 4 : - run ligandmpnn.py runs LigandMPNN on CPU CUDA VISIBLE DEVICES="" and shards pending designs across --num workers parallel CPU processes default: min 16, os.cpu count , using the same worker pool to write resequenced mmCIF structures in parallel. - evaluate design.py evaluates folded designs in parallel across up to min 24, os.cpu count CPU worker processes when --num workers=1 . - Manual / Multi-Node Sharding : You can also shard Stages 1, 3, or 4 explicitly across separate jobs or nodes by passing --worker id=<0..N-1 and --num workers=<N directly to run generator.py , run alphafold.py , or evaluate design.py . Generates protein backbones and initial sequences from a manifest: python run generator.py \ --manifest=examples/kemp eliminase/kemp manifest.json \ --output dir=./samples | Key Flag | Description | |---|---| | --manifest | Required Path to design manifest JSON. | | --output dir | Directory where generated structures, sequences, and metadata are written. | | --model dir | Model weights directory defaults to ./models/apnovo generator . | | --resume | Skip already completed designs in --output dir default: true . | | --return all data | Include unrolled diffusion trajectory diagnostics default: false . | | --worker id | 0-indexed worker ID for multi-GPU sharding default: 0 . | | --num workers | Total number of parallel workers for sharding default: 1 ; auto-spawned across visible GPUs when 1 . | Run python run generator.py --help for all options. Resequences generated backbones with LigandMPNN from within the primary environment: python run ligandmpnn.py \ --input dir=./samples \ --output dir=./samples \ --ligandmpnn dir=/path/to/LigandMPNN \ --python executable="/path/to/ligandmpnn/.venv-ligandmpnn/bin/python" | Key Flag | Description | |---|---| | --input dir | Directory holding designs to resequence located via design metadata . | | --output dir | Destination directory for resequenced structures and FASTA files. | | --ligandmpnn dir | Path to cloned LigandMPNN repository. | | --python executable | Python interpreter from the ligandmpnn environment. | | --temperature | Sampling temperature for sequence generation default: 0.1 . | | --checkpoint | Model checkpoint default: ligandmpnn v 32 030 25.pt . | | --ligand mpnn use side chain context | Enables side-chain context for fixed residues resequence.use side chain context : true by default . | | --num workers | Number of parallel CPU workers for LigandMPNN inference and structure post-processing default: min 16, os.cpu count . | Run python run ligandmpnn.py --help for all options. Predicts structures using AlphaFold 3. Folding runs against the AF3-LA weights with --fix standalone glycans=true , so that leaving atoms on covalent intermediates and unbonded glycan ligands are modelled rather than stripped: python run alphafold.py \ --input dir=./samples \ --output dir=./folded outputs \ --model dir=./models/af3 la | Key Flag | Description | |---|---| | --input dir | Directory containing designs to fold. | | --output dir | Destination directory for predicted structures and confidences. | | --model dir | Directory containing AlphaFold 3 weights defaults to models/af3 la , holding af3 leaving atom.bin.zst . | | --fix standalone glycans | Preserve leaving atoms on unbonded "standalone" glycan ligands default: true . | | --input structure | Structure to fold: generated default or resequenced . | | --seed | Random seed for AlphaFold 3 inference default: 230 . | | --resume | Skip designs already folded into --output dir default: true . | | --worker id | 0-indexed worker ID for multi-GPU sharding default: 0 . | | --num workers | Total number of parallel workers for sharding default: 1 ; auto-spawned across visible GPUs when 1 . | Run python run alphafold.py --help for all options. Computes various structural, geometric, and self-consistency metrics: python evaluate design.py \ --input dir=./folded outputs \ --output dir=./eval outputs | Key Flag | Description | |---|---| | --input dir | Directory containing folded structures and metadata. Repeatable. | | --output dir | Directory where <prefix evaluation.json and evaluation summary.csv will be written. | | --eval reference cif | Optional path to ground-truth motif CIF overrides metadata . | | --resume | Reuse evaluation for designs already in --output dir default: true . | | --worker id | 0-indexed worker ID when sharding across multiple evaluation jobs default: 0 . | | --num workers | Total number of shard workers when sharding across multiple evaluation jobs default: 1 ; single-worker runs auto-parallelize across up to min 24, os.cpu count CPU processes . | Run python evaluate design.py --help for all options. evaluate design.py computes key metrics across each folded state see metrics/ for complete implementations : - Self-Consistency : - rmsd :$C \alpha$ RMSD between the predicted model and designed scaffold Å . - tm score : Template Modeling score between predicted and designed structures 0 to 1 . - lddt & gdt ha : Local Distance Difference Test and High-Accuracy Global Distance Test scores. - - Catalytic Motif Preservation : - motif allatom rmsd : All-atom RMSD of catalytic motif residues with side-chain permutation symmetry. - motif bb aligned allatom rmsd : All-atom motif RMSD after aligning active-site backbones. - - Ligand Pocket Geometry : - mean pocket bb aligned ligand rmsd : Ligand RMSD after aligning pocket backbone atoms. Reference-dependent ligand consistency metrics mean pocket bb aligned ligand rmsd , pocket bb aligned ligand rmsd/<ligand , mean pocket bb rmsd , motif with ligand allatom rmsd , and interface lddt are only computed for folding states whose ligand heavy-atom set chain id, res id, res name, atom name and intra-residue covalent bond graph match the ligand in the generated structure; they are omitted for states folded with different ligands or atom-numbering topologies e.g. cleaved intermediates aei / ti2 or 4MU-Ac es . When a non-covalent substrate state and covalent transition state share the same heavy-atom names and bond connectivity such as DEHP esterase es vs. ti1 , these ligand consistency metrics are computed for both. - percent ligand bb clashes 1 5 : Percentage of ligand atoms clashing with protein backbone < 1.5 Å . Computed for all ligand-containing folding states. - - AlphaFold 3 Confidence : - plddt : Mean per-atom predicted lDDT 0 to 100 . - ptm & iptm : Predicted TM-score and Interface predicted TM-score. - Any publication that discloses findings arising from using this source code, the model parameters, or outputs produced by those should cite: @article{Wu2026, author = {Wu, Zachary and Abramson, Joshua and Frerix, Thomas and Chu, Alexander E. and Zhang, Ruijie K. and Schulz, Luca and Danson, Amy E. and Kwan, Tristan O. C. and Li, Wenliang K. and Kelly, Jacob and Li, Zi-Qi and Schneider, Rosalia G. and Thillaisundaram, Ashok and Patani, Harshnira and Zambaldi, Vinicius F. and Singh, Sukhdeep and La, David and Domecillo, Masy and Mora, Ariane N. and Reisenbauer, Julia C. and Zhang, Yu and Papa, Eliseo and Žemgulytė, Akvilė and Wu, Yu-Han and Žídek, Augustin and Shi, Jiaxin and Margand, Grace and Assem, Naila and Stephen, Kate and Emrich, Charlie and Liu, Peng and Colwell, Lucy and Hassabis, Demis and Fergus, Rob and Arnold, Frances H. and Kohli, Pushmeet and Wang, Jue}, title = {Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo}, journal = {bioRxiv}, year = {2026}, elocation-id = {2026.10.01.756017}, doi = {10.64898/2026.10.01.756017}, URL = {https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1}, eprint = {https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1.full.pdf} } Copyright 2026 Google LLC All software is licensed under the Apache License, Version 2.0 Apache 2.0 ; you may not use this file except in compliance with the Apache 2.0 license. You may obtain a copy of the Apache 2.0 license at: https://www.apache.org/licenses/LICENSE-2.0 https://www.apache.org/licenses/LICENSE-2.0 The AlphaProtein Novo Generator model parameters are made available under the AlphaProtein Novo Generator Model Parameters Terms of Use https://github.com/google-deepmind/alphaprotein-novo/blob/main/WEIGHTS TERMS OF USE.md the "APN Terms" ; you may not use these except in compliance with the Terms. You may obtain a copy of the Terms at https://github.com/google-deepmind/alphaprotein-novo/blob/main/WEIGHTS TERMS OF USE.md https://github.com/google-deepmind/alphaprotein-novo/blob/main/WEIGHTS TERMS OF USE.md . Any pre-computed outputs of AlphaProtein Novo Generator in this repository will be subject to the AlphaProtein Novo Generator Outputs Terms of Use https://github.com/google-deepmind/alphaprotein-novo/blob/main/OUTPUT TERMS OF USE.md the “Output Terms” , you may not use any output except in compliance with the Output Terms. You may obtain a copy of the Output Terms at https://github.com/google-deepmind/alphaprotein-novo/blob/main/OUTPUT TERMS OF USE.md https://github.com/google-deepmind/alphaprotein-novo/blob/main/OUTPUT TERMS OF USE.md . The AlphaFold 3 Leaving Atom model parameters are made available under the AlphaFold 3 Model Parameters Terms of Use https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS TERMS OF USE.md the "Terms" ; you may not use these except in compliance with the Terms. You may obtain a copy of the Terms at https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS TERMS OF USE.md https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS TERMS OF USE.md . All other materials are licensed under the Creative Commons Attribution 4.0 International License CC-BY . You may obtain a copy of the CC-BY license at: https://creativecommons.org/licenses/by/4.0/legalcode https://creativecommons.org/licenses/by/4.0/legalcode Unless required by applicable law or agreed to in writing, all software and materials distributed here under the Apache 2.0, the APN Terms, the Output Terms, the Terms or CC-BY licenses are distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the licenses for the specific language governing permissions and limitations under those licenses. You are solely responsible for determining the appropriateness of using the software, model parameters, materials or using or distributing outputs, and assume any and all risks associated with such use or distribution and your exercise of rights and obligations under these Terms. You and anyone you share output with are solely responsible for these and their subsequent uses. Output are predictions with varying levels of confidence and should be interpreted carefully. Use discretion before relying on, publishing, downloading or otherwise using AlphaProtein Novo Generator or AlphaFold 3 Leaving Atom. All software, model parameters, materials and any outputs you create are for theoretical modeling only. They are not intended, validated, or approved for clinical use. You should not use the software, model parameters, materials or outputs for clinical purposes or rely on them for medical or other professional advice. Any content regarding those topics is provided for informational purposes only and is not a substitute for advice from a qualified professional. This is not an official Google product.