Basic Usage

Workflow

Usage

A Guide to Rosetta Energy Calculation

Directory Structure

The Rosetta project directory is organized in a hierarchical manner to facilitate efficient workflow and computation. Below is an illustrative structure:

Rosetta Project Directory Structure

rosetta/
├── pdb/                      # Directory for storing original PDB files
├── pdb_list/                # Task list file directory
├── pdb_rosetta/             # Directory for storing calculation results
├── features.xml             # Rosetta feature definition file
├── rosetta_feature.sh       # Main script
├── rosetta_feature_sub.sh   # Sub-task scheduling script
└── rosetta_feature_parallel.sh  # Parallel computing script

Single - Task Calculation

Follow these steps to perform a single - task calculation:

# Go to the working directory
cd rosetta

# 1. Put the PDB file into the Pdb/ directory
# 2. Edit the task list pdb_list.txt and save it to pdb_list/

# Run single - task calculation
bash rosetta_feature.sh

Parallel Computing

For parallel computing, follow these steps:

cd rosetta

# 1. Split the task list (n is the number of lines per sub - task file)
cd pdb_list
split -l n -d -a 3 --additional-suffix=.txt pdb_list.txt list_part_
cd ..
# 2. Submit parallel tasks
bash rosetta_feature_sub.sh

Rosetta feature include

Name	Type	Description
total_score	float	Total score of the protein
dslf_fa13	float	Long-range interaction score
fa_atr	float	Attractive Van der Waals interaction score
fa_dun	float	Disulfide bond penalty term
fa_elec	float	Charge interaction score
fa_intra_rep	float	Intramolecular repulsion score
fa_intra_sol_xover4	float	Intramolecular solvation crossover term score
fa_rep	float	Intermolecular repulsion score
fa_sol	float	Solvation score
hbond_bb_sc	float	Backbone-sidechain hydrogen bond score
hbond_lr_bb	float	Long-range backbone hydrogen bond score
hbond_sc	float	Sidechain hydrogen bond score
hbond_sr_bb	float	Short-range backbone hydrogen bond score
lk_ball_wtd	float	Spherical weighted connectivity score
omega	float	Dihedral angle constraint score
p_aa_pp	float	Amino acid preference score
pro_close	float	Proline closure score
rama_prepro	float	Ramachandran map preprocessing score
ref	float	Reference score
yhh_planarity	float	Aromatic ring planarity score
cavity_volume	float	Cavity volume of the protein
contacts	float	Number of contacts
degree	float	Degree of connectivity
exposed_hydrophobics	float	Exposed hydrophobic residues count
exposed_polars	float	Exposed polar residues count
exposed_total	float	Total exposed residues count
net_charge	float	Net charge
nres	int	Number of residues
packstat	float	Burial statistics score
unsat_hbond2	float	Unsaturated hydrogen bond score

Data Preparation and Training/Inference Modules

Data Preparation Module

This module is designed to prepare your data for further processing and training. It organizes and processes raw PDB files, embeddings, and labels into a structured format suitable for model training.

Command Syntax

python ./access/data/data_prepare.py \
  --pdb_dir /path/to/pdb_files \
  --include_dir /path/to/include_files \
  --embedding_dir /path/to/embedding_output \
  --processed_dir /path/to/processed_output \
  --label_file labels.csv \
  [--split_mode STRATEGY] \
  [--num_workers NUM] \
  [--seed SEED]

Mandatory Arguments

Argument	Description
--pdb_dir	Directory containing input PDB files
--include_dir	Directory containing filter/label files
--embedding_dir	Base output directory for generated embeddings
--processed_dir	Output directory for processed datasets
--label_file	CSV file containing data labels

Optional Parameters

Argument	Default Value	Description
--split_mode	all_test	Data splitting strategy: train_split (8:1:1) all_train all_valid all_test
--num_workers	64	Number of parallel workers
--seed	42	Random seed for splitting

Training Module

The training module is responsible for training the model on the prepared dataset. It leverages the processed data to learn patterns and relationships within the protein structures.

Command Syntax

python ./access/training/train.py \
  --root_dir ./access/gvp_protein_embedding/processed \
  --resultFolder ./access/result \
  [--batch_size SIZE] \
  [--num_workers NUM]

Parameters

Argument	Default Value	Description
--root_dir	Required	Root directory of processed datasets
--resultFolder	Required	Output directory for models and results
--batch_size	8	Training batch size
--num_workers	2	Data loader workers

Inference Module

Once the model is trained, the inference module allows you to make predictions on new data. It uses the trained model to generate predictions based on the input data.

Command Syntax

python ./access/inference/inference.py \
  --model ./access/saved_models/best_model.pt \
  --root_dir ./access/gvp_protein_embedding/processed \
  --resultFolder ./access/result \
  [--print_true_label] \
  [--print_embedding] \
  [--batch_size SIZE] \
  [--num_workers NUM]

Mandatory Arguments

Argument	Description
--model	Path to trained model file (.pt)
--root_dir	Root directory of processed datasets
--resultFolder	Output directory for inference results

Optional Flags

Argument	Description
--print_true_label	Display true labels in output
--print_embedding	Output embeddings in results

Optional Parameters

Argument	Default Value	Description
--batch_size	8	Inference batch size
--num_workers	2	Data loader workers

Usage Notes

1. File Path Requirements:

   • All directory arguments require absolute paths or paths relative to execution context

   • Ensure write permissions for output directories

2. Performance Recommendations:

   • Set --num_workers to 2 for GPU and 16 for CPU.

   • Use --batch_size with 8 for GPU and 64 for CPU.