### Install ModelAngelo and Download Weights

Source: https://context7.com/3dem/model-angelo/llms.txt

Clone the repository, install ModelAngelo using the provided script, activate the conda environment, and verify the installation. For shared clusters, pre-download weights to a shared path.

```bash
# Clone the repository
git clone https://github.com/3dem/model-angelo.git
cd model-angelo

# Install (creates conda env named 'model_angelo', installs PyTorch + package)
source install_script.sh

# Activate
conda activate model_angelo

# Verify
model_angelo --version
# ModelAngelo 1.0.16

# For shared clusters: pre-download weights to a shared path
export TORCH_HOME=/public/model_angelo_weights
source install_script.sh --download-weights
```

--------------------------------

### Download and Install Model Weights

Source: https://context7.com/3dem/model-angelo/llms.txt

Downloads and installs model weights for a specified model type (e.g., 'nucleotides'). Returns the path where the weights were installed.

```python
from model_angelo.utils.torch_utils import download_and_install_model

bundle_path = download_and_install_model("nucleotides")
print(f"Weights installed at: {bundle_path}")
```

--------------------------------

### Install ModelAngelo with Downloaded Weights

Source: https://github.com/3dem/model-angelo/blob/main/README.md

For shared computational environments, install ModelAngelo and download weights to a specified public directory. Ensure TORCH_HOME is set correctly.

```bash
export TORCH_HOME=/public/model_angelo_weights
cd model-angelo
source install_script.sh --download-weights
```

--------------------------------

### Verify ModelAngelo Installation

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Check if the installation was successful by viewing the help message for the 'build' command. This confirms that ModelAngelo is runnable.

```bash
model_angelo build -h
```

--------------------------------

### Check Conda Installation

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Verify that Conda is installed and accessible in your environment. This is a prerequisite for installing ModelAngelo.

```bash
conda info
```

--------------------------------

### Run ModelAngelo Installation Script

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Execute the installation script to set up the ModelAngelo conda environment. This script handles dependencies and environment creation.

```bash
cd model-angelo
source install_script.sh
```

--------------------------------

### Clone ModelAngelo Repository

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Download the ModelAngelo source code from its GitHub repository. This is the first step in the installation process.

```bash
git clone https://github.com/3dem/model-angelo.git
```

--------------------------------

### Pre-download Model Angelo weights

Source: https://context7.com/3dem/model-angelo/llms.txt

Download and cache model bundle weights to `$TORCH_HOME/checkpoints/model_angelo_v1.0/<bundle>`. This is useful for cluster setups where compute nodes lack internet access.

```bash
export TORCH_HOME=/shared/weights
```

```bash
model_angelo setup_weights --bundle-name nucleotides
```

```bash
model_angelo setup_weights --bundle-name nucleotides_no_seq
```

--------------------------------

### model_angelo setup_weights

Source: https://context7.com/3dem/model-angelo/llms.txt

Pre-download and cache model bundle weights. This is useful for cluster environments where compute nodes may not have direct internet access.

```APIDOC
## `model_angelo setup_weights` — Pre-download model weights

Download and cache model bundle weights to `$TORCH_HOME/checkpoints/model_angelo_v1.0/<bundle>`. Useful for cluster setups where compute nodes lack internet access.

### Usage
```bash
model_angelo setup_weights --bundle-name <bundle_name>
```

### Parameters
- **--bundle-name**: The name of the model bundle to download. Available options:
    - `nucleotides` (default for build)
    - `nucleotides_no_seq` (default for build_no_seq)
    - `original` (legacy protein-only)
    - `original_no_seq` (legacy protein-only, no seq)
    - `small_gpu` (for GPUs with < 8 GB VRAM)
```

--------------------------------

### Programmatic model building with sequences (Python API)

Source: https://context7.com/3dem/model-angelo/llms.txt

Use the Python API for building models with sequences. Requires specifying volume and FASTA file paths, and an output directory. The `build_main` function handles the process.

```python
import os
from types import SimpleNamespace

from model_angelo.apps.build import main as build_main

args = SimpleNamespace(
    volume_path="map.mrc",
    protein_fasta="protein.fasta",
    rna_fasta=None,
    dna_fasta=None,
    output_dir="output",
    mask_path=None,
    device=None,                        # auto-select GPU
    config_path=None,
    model_bundle_name="nucleotides",
    model_bundle_path=None,
    keep_intermediate_results=False,
    pipeline_control="",
)
build_main(args)
```

--------------------------------

### Build Model with Known Sequences (with Mask and GPU)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build an atomic model from a cryo-EM map with known sequences, specifying an explicit solvent mask and selecting a GPU for computation. This command performs C-alpha prediction and GNN refinement.

```bash
# With an explicit solvent mask and GPU selection
model_angelo build \
    -v map.mrc \
    -pf protein.fasta \
    -m mask.mrc \
    --device 0 \
    -o my_output
```

--------------------------------

### Build Model with Known Sequences (Multi-GPU)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build an atomic model from a cryo-EM map with known sequences using multiple GPUs for accelerated computation. This command leverages C-alpha prediction and GNN refinement.

```bash
# Multi-GPU (GPUs 0 and 1)
model_angelo build -v map.mrc -pf protein.fasta --device 0,1 -o my_output
```

--------------------------------

### Build Model with Known Sequences (Protein, RNA, DNA)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build an atomic model from a cryo-EM map including known protein, RNA, and DNA sequences. This command utilizes C-alpha prediction and GNN refinement, outputting various model and score files.

```bash
# With RNA and DNA sequences included
model_angelo build \
    -v map.mrc \
    -pf protein.fasta \
    -rf rna.fasta \
    -df dna.fasta \
    -o my_output
```

--------------------------------

### Build Model with FASTA Sequence

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Recommended for cryo-EM maps with resolutions exceeding 4 Å and available FASTA sequences. Specify protein, DNA, and RNA sequences in separate files if applicable.

```bash
model_angelo build -v map.mrc -pf prot.fasta -o output
```

```bash
model_angelo build -v map.mrc -pf prot.fasta -df dna.fasta -rf rna.fasta -o output
```

--------------------------------

### Build Model with Known Sequences (Protein Only)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build an atomic model from a cryo-EM map using known protein sequences. This command runs C-alpha prediction followed by three rounds of GNN refinement and outputs pruned and raw mmCIF files, along with entropy score files.

```bash
# Minimal: protein-only map
model_angelo build \
    -v map.mrc \
    -pf sequences.fasta \
    -o my_output
```

--------------------------------

### Build Model without Known Sequences (with Mask and GPU)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build a backbone model from a cryo-EM map without known sequences, specifying a solvent mask and selecting a GPU. This command utilizes a sequence-free GNN and generates HMM profiles.

```bash
# With mask and specific GPU
model_angelo build_no_seq \
    -v map.mrc \
    -m mask.mrc \
    --device 1 \
    -o no_seq_output
```

--------------------------------

### Create Shared ModelAngelo Bash Script

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Provide a bash script for users to easily activate the ModelAngelo environment and run the program. This simplifies access on shared clusters.

```bash
#!/bin/bash
source `which activate` model_angelo
model_angelo "$@"
```

--------------------------------

### Build Model with Known Sequences (Keep Intermediate Results)

Source: https://context7.com/3dem/model-angelo/llms.txt

Build an atomic model from a cryo-EM map with known sequences, retaining intermediate C-alpha and per-round GNN outputs for debugging purposes. This command performs the standard C-alpha prediction and GNN refinement pipeline.

```bash
# Keep intermediate C-alpha and per-round GNN outputs for debugging
model_angelo build -v map.mrc -pf protein.fasta -o my_output \
    --keep-intermediate-results
```

--------------------------------

### Programmatic model building without sequences (Python API)

Source: https://context7.com/3dem/model-angelo/llms.txt

Use the Python API for building models without sequences. Requires specifying volume and output directory. The `build_no_seq_main` function handles the process, and can output HMM profiles.

```python
from model_angelo.apps.build_no_seq import main as build_no_seq_main

args_no_seq = SimpleNamespace(
    volume_path="map.mrc",
    output_dir="no_seq_out",
    mask_path=None,
    device="0",
    config_path=None,
    model_bundle_name="nucleotides_no_seq",
    model_bundle_path=None,
    keep_intermediate_results=False,
    pipeline_control="",
)
build_no_seq_main(args_no_seq)
```

--------------------------------

### Add Miniconda to PATH

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Manually add the miniconda3 bin directory to your PATH environment variable if the 'activate' binary is not found. This resolves 'Binary activate not found' errors.

```bash
export PATH="$PATH:/path/to/miniconda3/bin"
```

--------------------------------

### Build Model without Known Sequences

Source: https://context7.com/3dem/model-angelo/llms.txt

Build a backbone model from a cryo-EM map when sequences are unknown. This command uses a sequence-free GNN and generates per-chain HMM profiles for subsequent database searching.

```bash
# Basic usage
model_angelo build_no_seq \
    -v map.mrc \
    -o no_seq_output
```

--------------------------------

### `model_angelo build`

Source: https://context7.com/3dem/model-angelo/llms.txt

Automated model building with known sequences. This command builds an atomic model from a cryo-EM map when protein, RNA, or DNA sequences are provided. It outputs pruned and raw mmCIF files, along with per-residue entropy scores.

```APIDOC
## `model_angelo build` — Automated model building with known sequences

Build an atomic model from a cryo-EM map when protein (and optionally RNA/DNA) sequences are known. Runs C-alpha prediction followed by three rounds of GNN refinement. Outputs a pruned mmCIF file, a raw mmCIF file, and per-residue entropy score files.

### Method
`model_angelo build`

### Parameters
#### Path Parameters
- **-v** (string) - Required - Path to the input cryo-EM map (.mrc file).
- **-pf** (string) - Required - Path to the protein sequence FASTA file.
- **-rf** (string) - Optional - Path to the RNA sequence FASTA file.
- **-df** (string) - Optional - Path to the DNA sequence FASTA file.
- **-m** (string) - Optional - Path to an explicit solvent mask (.mrc file).
- **-o** (string) - Required - Output directory for generated files.

#### Query Parameters
- **--device** (integer or comma-separated list) - Optional - GPU device ID(s) to use (e.g., 0 or 0,1).
- **--keep-intermediate-results** - Optional - Flag to keep intermediate C-alpha and per-round GNN outputs for debugging.

### Request Example
```bash
# Minimal: protein-only map
model_angelo build \
    -v map.mrc \
    -pf sequences.fasta \
    -o my_output

# With RNA and DNA sequences included
model_angelo build \
    -v map.mrc \
    -pf protein.fasta \
    -rf rna.fasta \
    -df dna.fasta \
    -o my_output

# With an explicit solvent mask and GPU selection
model_angelo build \
    -v map.mrc \
    -pf protein.fasta \
    -m mask.mrc \
    --device 0 \
    -o my_output

# Multi-GPU (GPUs 0 and 1)
model_angelo build -v map.mrc -pf protein.fasta --device 0,1 -o my_output

# Keep intermediate C-alpha and per-round GNN outputs for debugging
model_angelo build -v map.mrc -pf protein.fasta -o my_output \
    --keep-intermediate-results
```

### Expected output files:
- `my_output/my_output.cif` (pruned final model)
- `my_output/my_output_raw.cif` (raw, unpruned model)
- `my_output/entropy_scores/my_output.cif`
- `my_output/model_angelo.log`
```

--------------------------------

### Python API: build

Source: https://context7.com/3dem/model-angelo/llms.txt

Programmatically build a model from a cryo-EM map and sequence information using the `build` function.

```APIDOC
```python
from types import SimpleNamespace
from model_angelo.apps.build import main as build_main

args = SimpleNamespace(
    volume_path="map.mrc",
    protein_fasta="protein.fasta",
    rna_fasta=None,
    dna_fasta=None,
    output_dir="output",
    mask_path=None,
    device=None,  # auto-select GPU
    config_path=None,
    model_bundle_name="nucleotides",
    model_bundle_path=None,
    keep_intermediate_results=False,
    pipeline_control="",
)
build_main(args)
# Output: output/output.cif
```
```

--------------------------------

### Specify GPU Device for ModelAngelo

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Control which GPU(s) ModelAngelo utilizes by specifying their IDs with the --device flag. Supports single or multiple GPU selections.

```bash
model_angelo --device 0
```

```bash
model_angelo --device 0,1
```

--------------------------------

### Search HMM Profiles with Adjusted Sensitivity

Source: https://context7.com/3dem/model-angelo/llms.txt

Search per-chain HMM profiles against a reference database with adjusted sensitivity thresholds for more stringent or relaxed matching. Lower E-value and F3-score thresholds lead to more stringent searches.

```bash
# Adjust sensitivity thresholds (lower E = more stringent)
model_angelo hmm_search \
    -i no_seq_output \
    -f UP000005640_9606.fasta \
    --E 1e-3 \
    --F3 1e-8 \
    -o hmm_strict_output
```

--------------------------------

### Search HMM Profiles against Protein Database

Source: https://context7.com/3dem/model-angelo/llms.txt

Search the per-chain HMM profiles generated by `build_no_seq` against a reference FASTA database, such as a UniProt reference proteome. This command outputs per-chain `.hhr` hit files and aggregated CSV summaries.

```bash
# Download example proteome (human)
wget https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/
reference_proteomes/Eukaryota/UP000005640/UP000005640_9606.fasta.gz
gunzip UP000005640_9606.fasta.gz

# Search protein chains against the proteome
model_angelo hmm_search \
    -i no_seq_output \
    -f UP000005640_9606.fasta \
    -o hmm_search_output
```

--------------------------------

### Perform HMM Search for Protein Identification

Source: https://github.com/3dem/model-angelo/blob/main/README.md

After building a model without sequences, use this command to search for protein identities using generated HMM profiles and a downloaded FASTA database. Results are saved in HMMER3 format.

```bash
model_angelo hmm_search --i output --f PATH_TO_DB --o hmm_output
```

--------------------------------

### Build Model without FASTA Sequence

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Use when protein sequences are unknown. This version provides HMM profile files for subsequent database searching.

```bash
model_angelo build_no_seq -v map.mrc -o output
```

--------------------------------

### Refine an existing model with Model Angelo

Source: https://context7.com/3dem/model-angelo/llms.txt

Use this command to perform a single round of GNN refinement on a pre-built PDB or mmCIF structure. It can optionally emit HMM profiles for downstream sequence searches.

```bash
model_angelo refine \
    -i existing_model.cif \
    -v map.mrc \
    -o refined_output
```

```bash
model_angelo refine \
    -i existing_model.cif \
    -v map.mrc \
    -w \
    -o refined_with_hmm
```

--------------------------------

### Python API: build_no_seq

Source: https://context7.com/3dem/model-angelo/llms.txt

Programmatically build a model from a cryo-EM map without providing sequence information using the `build_no_seq` function.

```APIDOC
```python
from types import SimpleNamespace
from model_angelo.apps.build_no_seq import main as build_no_seq_main

args_no_seq = SimpleNamespace(
    volume_path="map.mrc",
    output_dir="no_seq_out",
    mask_path=None,
    device="0",
    config_path=None,
    model_bundle_name="nucleotides_no_seq",
    model_bundle_path=None,
    keep_intermediate_results=False,
    pipeline_control="",
)
build_no_seq_main(args_no_seq)
# Output: no_seq_out/no_seq_out.cif
#         no_seq_out/hmm_profiles/*.hmm
```
```

--------------------------------

### Search HMM Profiles against RNA Database

Source: https://context7.com/3dem/model-angelo/llms.txt

Search per-chain HMM profiles against an RNA reference database. This command is similar to protein HMM searching but requires specifying the alphabet as RNA and outputs per-chain hit reports and CSV summaries.

```bash
# Search RNA chains (change --alphabet accordingly)
model_angelo hmm_search \
    -i no_seq_output \
    -f rna_database.fasta \
    --alphabet RNA \
    -o hmm_rna_output
```

--------------------------------

### Activate ModelAngelo Conda Environment

Source: https://github.com/3dem/model-angelo/blob/main/README.md

Activate the created conda environment to use ModelAngelo. This command makes the ModelAngelo executables available in your current shell session.

```bash
conda activate model_angelo
```

--------------------------------

### `model_angelo build_no_seq`

Source: https://context7.com/3dem/model-angelo/llms.txt

Model building without known sequences. This command builds a backbone model when sequences are unknown. It generates per-chain HMM profiles in HMMER3 format, which can be used for sequence identification against proteome databases.

```APIDOC
## `model_angelo build_no_seq` — Model building without known sequences

Build a backbone model when sequences are unknown. Uses a sequence-free GNN and additionally writes per-chain HMM profiles in HMMER3 format that can be searched against proteome databases to identify the chains.

### Method
`model_angelo build_no_seq`

### Parameters
#### Path Parameters
- **-v** (string) - Required - Path to the input cryo-EM map (.mrc file).
- **-m** (string) - Optional - Path to an explicit solvent mask (.mrc file).
- **-o** (string) - Required - Output directory for generated files.

#### Query Parameters
- **--device** (integer) - Optional - GPU device ID to use.

### Request Example
```bash
# Basic usage
model_angelo build_no_seq \
    -v map.mrc \
    -o no_seq_output

# With mask and specific GPU
model_angelo build_no_seq \
    -v map.mrc \
    -m mask.mrc \
    --device 1 \
    -o no_seq_output
```

### Expected output files:
- `no_seq_output/no_seq_output.cif` (built model)
- `no_seq_output/no_seq_output_entropy_score.cif`
- `no_seq_output/hmm_profiles/A_u.hmm` (one .hmm file per chain)
- `no_seq_output/hmm_profiles/B_u.hmm`
- `no_seq_output/model_angelo.log`
```

--------------------------------

### Per-residue evaluation with Model Angelo

Source: https://context7.com/3dem/model-angelo/llms.txt

Generate a residue-level breakdown of various quality metrics, saved as a tab-separated table. Can also load a saved table to plot metrics vs. B-factor or filter by match type.

```bash
model_angelo eval_per_resid \
    -p predicted.cif \
    -t ground_truth.cif \
    -o per_residue_report.tsv
```

```bash
model_angelo eval_per_resid \
    -d per_residue_report.tsv \
    --plot
```

```bash
model_angelo eval_per_resid \
    -p predicted.cif \
    -t ground_truth.cif \
    --match-type nucleotide \
    -o nuc_report.tsv
```

--------------------------------

### `model_angelo hmm_search`

Source: https://context7.com/3dem/model-angelo/llms.txt

Sequence identification via HMM search. This command searches the per-chain HMM profiles generated by `build_no_seq` against a reference FASTA database. It outputs per-chain HMMER hit report files (.hhr) and aggregated CSV summaries.

```APIDOC
## `model_angelo hmm_search` — Sequence identification via HMM search

Search the per-chain HMM profiles produced by `build_no_seq` against a reference FASTA database (e.g., a UniProt reference proteome). Outputs per-chain `.hhr` hit files and aggregated CSV summaries.

### Method
`model_angelo hmm_search`

### Parameters
#### Path Parameters
- **-i** (string) - Required - Input directory containing HMM profiles (output from `build_no_seq`).
- **-f** (string) - Required - Path to the reference FASTA database file.
- **-o** (string) - Required - Output directory for search results.

#### Query Parameters
- **--alphabet** (string) - Optional - Alphabet type for the sequences (e.g., 'PROTEIN', 'RNA'). Defaults to 'PROTEIN'.
- **--E** (float) - Optional - E-value threshold for reporting hits. Lower values are more stringent.
- **--F3** (float) - Optional - Bit score threshold for reporting hits. Lower values are more stringent.

### Request Example
```bash
# Download example proteome (human)
wget https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/reference_proteomes/Eukaryota/UP000005640/UP000005640_9606.fasta.gz
gunzip UP000005640_9606.fasta.gz

# Search protein chains against the proteome
model_angelo hmm_search \
    -i no_seq_output \
    -f UP000005640_9606.fasta \
    -o hmm_search_output

# Search RNA chains (change --alphabet accordingly)
model_angelo hmm_search \
    -i no_seq_output \
    -f rna_database.fasta \
    --alphabet RNA \
    -o hmm_rna_output

# Adjust sensitivity thresholds (lower E = more stringent)
model_angelo hmm_search \
    -i no_seq_output \
    -f UP000005640_9606.fasta \
    --E 1e-3 \
    --F3 1e-8 \
    -o hmm_strict_output
```

### Expected output files:
- `hmm_search_output/A.hhr` (per-chain hit report)
```

--------------------------------

### Read and Process FASTA Sequences

Source: https://context7.com/3dem/model-angelo/llms.txt

Reads sequences and names from a FASTA file, then prints details for each sequence. Also calculates and prints the total number of residues across all sequences in the FASTA file.

```python
from model_angelo.utils.fasta_utils import read_fasta, fasta_to_unified_seq

sequences, names = read_fasta("protein.fasta")
for seq, name in zip(sequences, names):
    print(f"{name}: {len(seq.seq)} residues, chains={seq.chains}")

unified, total_len = fasta_to_unified_seq("protein.fasta")
print(f"Total residues across all sequences: {total_len}")
```

--------------------------------

### model_angelo evaluate

Source: https://context7.com/3dem/model-angelo/llms.txt

Compare a predicted model against a ground-truth structure to compute quality metrics such as RMSD, lDDT, recall, and precision.

```APIDOC
## `model_angelo evaluate` — Compare prediction to ground truth

Compute quality metrics (backbone RMSD, Cα RMSD, lDDT, recall, precision, sequence match) between a predicted mmCIF and a reference ground-truth mmCIF.

### Usage
```bash
model_angelo evaluate -p <predicted.cif> -t <ground_truth.cif> [--csv-format] [--name <sample_name>] [--match-type <protein|nucleotide>] [--output-file <path>] [--output-structure <path>]
```

### Parameters
- **-p, --prediction**: Path to the predicted model file (mmCIF).
- **-t, --target**: Path to the ground-truth model file (mmCIF).
- **--csv-format**: Optional flag to output results in CSV format.
- **--name**: Optional name for the sample, used with CSV format.
- **--match-type**: Optional. Specify the type of match to evaluate (`protein` or `nucleotide`). Defaults to all.
- **--output-file**: Optional path to save evaluation results to a file.
- **--output-structure**: Optional path to save an annotated structure file where B-factors represent correctness.
```

--------------------------------

### Evaluate Protein Model Fit Report

Source: https://context7.com/3dem/model-angelo/llms.txt

Generates a fit report comparing predicted and target protein structures. Requires specifying max distance, rounds of matching, match type, and an output structure file.

```python
from model_angelo.apps.evaluate import get_all_atom_fit_report
from model_angelo.utils.protein import get_protein_from_file_path

predicted = get_protein_from_file_path("output/output.cif")
target    = get_protein_from_file_path("ground_truth.cif")

backbone_rms, ca_rms, lddt, recall, precision, seq_match = get_all_atom_fit_report(
    predicted, target,
    max_dist=3.0,
    two_rounds=True,
    match_type="protein",          # "protein", "nucleotide", or "both"
    output_structure="annotated.cif",
)
print(f"Recall={recall:.3f}  Precision={precision:.3f}  lDDT={lddt:.3f}")
```

--------------------------------

### Compare prediction to ground truth with Model Angelo

Source: https://context7.com/3dem/model-angelo/llms.txt

Compute quality metrics between a predicted mmCIF and a reference ground-truth mmCIF. Supports human-readable, CSV, and pickle output formats, and can filter by match type (protein/nucleotide).

```bash
model_angelo evaluate \
    -p predicted.cif \
    -t ground_truth.cif
```

```bash
model_angelo evaluate \
    -p predicted.cif \
    -t ground_truth.cif \
    --csv-format \
    --name my_sample
```

```bash
model_angelo evaluate \
    -p predicted.cif \
    -t ground_truth.cif \
    --match-type protein
```

```bash
model_angelo evaluate \
    -p predicted.cif \
    -t ground_truth.cif \
    --output-file results/eval.log \
    --output-structure results/annotated.cif
```

--------------------------------

### model_angelo refine

Source: https://context7.com/3dem/model-angelo/llms.txt

Refine an existing model using a single round of GNN refinement. This can improve externally built models or poly-alanine chains and optionally emit HMM profiles.

```APIDOC
## `model_angelo refine` — Refine an existing model

Run a single round of GNN refinement on a pre-built PDB or mmCIF structure. Useful for improving an externally built model or poly-alanine chain. Can optionally emit HMM profiles.

### Usage
```bash
model_angelo refine -i <input_model.cif> -v <map.mrc> -o <output_dir> [-w]
```

### Parameters
- **-i, --input**: Path to the existing model file (PDB or mmCIF).
- **-v, --volume**: Path to the electron density map (MRC format).
- **-o, --output**: Directory to save the refined output.
- **-w, --write-hmm**: Optional flag to emit HMM profiles for downstream sequence search.
```

--------------------------------

### model_angelo eval_per_resid

Source: https://context7.com/3dem/model-angelo/llms.txt

Generate a residue-level breakdown of structural quality metrics, including RMSD, sequence match, and B-factor correlation.

```APIDOC
## `model_angelo eval_per_resid` — Per-residue evaluation

Produce a residue-level breakdown of Cα RMSD, backbone RMSD, all-atom RMSD, sequence match, and B-factor correlation, saved as a tab-separated table.

### Usage
```bash
model_angelo eval_per_resid -p <predicted.cif> -t <ground_truth.cif> -o <output.tsv> [--match-type <protein|nucleotide>] [--plot]
```

### Parameters
- **-p, --prediction**: Path to the predicted model file (mmCIF).
- **-t, --target**: Path to the ground-truth model file (mmCIF).
- **-o, --output**: Path to save the per-residue report (TSV format).
- **-d, --data**: Path to a previously saved per-residue TSV file for plotting.
- **--match-type**: Optional. Specify the type of residues to evaluate (`protein` or `nucleotide`).
- **--plot**: Optional flag to plot metrics from a provided data file.
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.