Outputs of ProMaya
The ProMaya pipeline generates a highly structured output directory. By default, all results are written to results/promaya_nf/ (configurable via the --outdir parameter).
Output Directory Structure
results/promaya_nf/
âââ master_predictions.tsv # 1. Primary aggregated results
âââ predictions/
â âââ <pair_id_1>/ # 2. Pair-specific results
â â âââ probability.tsv
â â âââ logs/
â â â âââ prediction.log
â â âââ gradcam/ # 3. Interpretability data (if --gradcam)
â â â âââ protein_a_atoms.tsv
â â â âââ protein_a_residue.tsv
â â â âââ protein_a_surface.tsv
â â â âââ interaction_pairs.tsv
â â â âââ top10_residue_correlation.tsv
â â âââ images/ # 4. 2D Visualizations (if --gradcam)
â â â âââ *_attribution.png
â â âââ structure/ # 5. 3D Visualizations (if --gradcam)
â â âââ *_structure_graph.png
â â âââ *_correlation_heatmap.png
â â âââ *_view.pml
â âââ <pair_id_N>/
âââ disorder/ # 6. Disorder scan results (if --run_disorder)
â âââ <protein_basename_1>/
â â âââ disorder_out/
â â âââ data/
â â â âââ *.tsv
â â âââ images/
â â âââ *.png
â âââ <protein_basename_N>/
âââ pipeline_info/ # 7. Nextflow provenance and reports
âââ execution_report_<timestamp>.html
âââ execution_timeline_<timestamp>.html
âââ execution_trace_<timestamp>.txt
âââ pipeline_dag_<timestamp>.svg
Detailed File Descriptions
1. master_predictions.tsv
This is the primary output file you should consult first. It is an aggregated table containing the predictions for every pair processed in the run, sorted by interaction_probability in descending order.
Columns:
- pair_id: The ID defined in your samplesheet.
- protein_a / protein_b: Paths to the input structures.
- checkpoint: The model checkpoint used.
- interaction_probability: The raw sigmoid output (0.0 to 1.0).
- threshold: The classification threshold used (default: 0.5).
- predicted_interaction: Binary classification (1 if probability >= threshold, else 0).
2. Pair-Specific Data (predictions/<pair_id>/)
probability.tsv: The individual prediction file for this specific pair.logs/prediction.log: Detailed execution log for the inference process, useful for debugging structural parsing issues (e.g., missing atoms, failed surface generation).
3. Interpretability Data (gradcam/)
Generated only if --gradcam true is active.
These TSV files map the Graph Grad-CAM attribution scores back to the original physical structures.
- protein_*_atoms.tsv: Attribution scores per atom. Identifies specific chemical groups driving the interaction.
- protein_*_residue.tsv: Attribution scores aggregated per amino acid residue.
- protein_*_surface.tsv: Attribution scores mapped to the generated solvent-accessible surface point cloud.
- interaction_pairs.tsv: Cross-protein attention scores highlighting which residues on Protein A strongly interact with which residues on Protein B.
4. 2D Visualizations (images/)
*_attribution.png: Static 2D plots summarizing the distribution of attribution scores across the protein sequence.
5. 3D Visualizations (structure/)
*_correlation_heatmap.png: A 2D heatmap matrix visualizing the cross-protein attention between the top interacting residues.*_view.pml: A dynamically generated PyMOL script. Opening this script in PyMOL will automatically load the PDB files and color the surfaces/ribbons according to the computed Grad-CAM attribution scores (red = high importance, blue = low importance).*_structure_graph.png: A rendered PNG of the structures colored by attribution (requires PyMOL to be available in the execution environment).
6. Disorder Analysis (disorder/)
Generated only if --run_disorder true is active.
- Contains raw TSV data and plotted PNG profiles of the intrinsic disorder probability across the sequence length of each unique protein, generated via IUPred2A/MetaPredict.
7. Pipeline Provenance (pipeline_info/)
Standard Nextflow reporting files containing execution times, memory/CPU utilization per process, and an interactive DAG visualization of the workflow.