PhosKinTime Data Preprocessing & Mapping

Note for contributors: The canonical version of this file is processing/README.md at the repository root. This docs copy mirrors it. Keep both in sync when updating documentation.

This workflow prepares and maps time-series data for kinase and transcription factor optimization models from raw proteomics and transcriptomics datasets.

Structure

phoskintime/
├── processing/
│   ├── cleanup.py     # Data cleaning and preparation
│   └── map.py         # Optimization result mapping and network table generation
├── raw/               # Input CSVs (CollecTRI, MS Gaussian, Rout Limma)
├── kinopt/data/       # Kinase model inputs
├── tfopt/data/        # TF model inputs
└── data/              # Network export for Cytoscape

Scripts Overview

cleanup.py

Performs the following steps:

1. TF-mRNA Interaction Cleanup

   • Filters complex interactions in CollecTRI
   • Keeps only TFs matching phospho-interactions in input2.csv

2. Proteomics Data Transformation

   • Transforms MS Gaussian predictions with 2^mean
   • Formats phosphorylation sites, saves to input1.csv

3. Error Propagation

   • Computes std propagation:
     σ_y = 2^x * ln(2) * σ_x
   • Saves to input1_wstd.csv

4. Transcriptomics Cleanup

   • Transforms Rout Limma values with 2^x
   • Saves to input3.csv

5. Gene Symbol Mapping

   • Replaces Ensembl/Entrez IDs with gene symbols (using MyGeneInfo)

6. File Management

   • Moves cleaned files to kinopt/data/ and tfopt/data/

map.py

This script processes optimization results for transcription factors (TFs) and kinases, mapping their interactions with mRNA and phosphorylation sites. It generates Cytoscape-compatible edge and node tables for network visualization.

Key Features:

• TF-mRNA Mapping: Extracts non-zero optimization results and groups mRNA by associated TFs and their strengths.
• Kinase-Phosphorylation Mapping: Maps kinases to mRNA and phosphorylation sites based on optimization results.
• Cytoscape Table Generation: Creates edge and node tables for network visualization, including interaction types and strengths.
• Kinetic Strength Integration: Adds kinetic strength columns to mapping files for further analysis.

Inputs

Place the following raw data in processing/raw/:

• CollecTRI.csv
• MS_Gaussian_updated_09032023.csv
• Rout_LimmaTable.csv
• input2.csv (phospho interactions)

Input File Schemas

CollecTRI.csv

TF–gene regulatory interaction database. cleanup.py uses these columns:

Other columns are ignored. Only interactions where Target appears in input2.csv's GeneID column are retained.

Common validation failures: - Rows where source starts with COMPLEX: — excluded automatically - NaN or blank source_genesymbol / target_genesymbol — excluded automatically - Duplicate rows — deduplicated automatically

MS_Gaussian_updated_09032023.csv

Mass-spectrometry phosphoproteomics data with Gaussian-fitted log2 abundances. cleanup.py uses these columns:

The script pivots the data to wide format using (GeneID, Psite) as index and unit_time as columns. After transformation: - predict_mean values are exponentiated: 2^predict_mean - predict_std is propagated: σ_y = 2^predict_mean * ln(2) * predict_std - Only rows with Psite starting with Y_, S_, T_, or empty are retained - Time columns are renamed x1–x14

Minimal example row:

GeneID,site,unit_time,predict_mean,predict_std
ABL2,S256,0,1.23,0.05
ABL2,S256,1,1.45,0.06

Rout_LimmaTable.csv

RNA time-series derived from Rout Limma differential expression analysis. cleanup.py uses these columns:

All 9 time-point columns are required. Values are exponentiated (2^x) to convert from log2 to linear fold-change. Renamed to x1–x9 in input3.csv.

Common validation failures: - Missing time-point columns — script will raise KeyError - Non-numeric values in time-point columns — will cause TypeError during 2^x computation

input2.csv

Kinase-substrate network / phospho-interaction metadata. Used both as raw input by cleanup.py and as a standard input by kinopt.

Additional columns may be present and are preserved in outputs.

Common validation failures: - Missing GeneID column — cleanup.py will raise KeyError - Empty GeneID values — filtered out automatically

Outputs

Notes

• Complex TF interactions (e.g. COMPLEX:TF1/TF2) are excluded.
• Kinase-only proteins not appearing in CollecTRI (e.g. PAK2) are excluded from TF mapping.
• Unmappable GeneIDs are printed at runtime.