Skip to content

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

From RNA-seq to a simulated protein–ligand complex: colorectal cancer and carbonic anhydrase VII

An end-to-end computational study that follows one gene from a transcriptomic signal to a structural hypothesis. Paired tumour/normal RNA-seq from colorectal cancer identifies the biology; functional enrichment interprets it; a ten-gene panel is locked and reproduced in an independent cohort; carbonic anhydrase VII (CA7) is carried into zinc-aware molecular docking; and the selected complex is tested for stability with molecular dynamics.

Every stage is documented — including the parts that failed, because those turned out to be the most informative.

Project overview

Click the picture to open it at full resolution.


Key results

Result
Genes tested 19,089 (6 patients, paired tumour vs normal)
Differentially expressed 5,038 at FDR < 0.05 → 2,854 with |log₂FC| ≥ 1 (894 up, 1,960 down)
Biology proliferation programmes up; differentiated colonic epithelium and transport lost
Independent validation ten-gene panel reproduced 10/10 in a separate 34-patient cohort (97.8 % agreement)
Structural target CA7 — log₂FC −5.06, padj 1.4 × 10⁻¹², 1.10 Å ligand-bound structure (PDB 8Q3U)
Standard docking plain AutoDock Vina failed redocking (10.07 Å) — a scoring failure, not a search failure
Zinc-aware docking only AutoDock4Zn + anionic sulfonamide recovered the zinc anchor — head group 0.785 Å, whole ligand 4.12 Å (partial validation)
20-compound screen Zn–N < 2.5 Å in 11/11 primary sulfonamides vs 0/9 controls; docking score did not separate the classes
Selected compound chlorthalidone (ZINC000000057255), Zn–N 1.93 Å vs 1.94 Å crystallographic — a known CA inhibitor
Molecular dynamics 6 ns, 41,271 atoms: Zn–N 2.214 ± 0.065 Å, four-coordinate in 100 % of frames, backbone RMSD 0.94 Å, 87.9 % of the ligand buried

Every number above is re-derived from the result files, with its source, in 04_Report/REPORT_DATA_CHECK.md.

The full write-up is the report: 04_Report/FINAL_CRC_CA7_PROJECT_REPORT.pdf.


The pipeline

1 · RNA-seq and functional enrichment — 01_RNAseq/

Raw counts for GSE50760 (primary tumour and matched normal colon from six patients) were analysed with DESeq2 using a patient-blocked design, ~ patient + condition, after filtering to genes with ≥ 10 counts in ≥ 6 samples. The patient pairing was not in the supplied metadata and was recovered from the GEO sample titles. Enrichment was run separately on up- and down-regulated genes (GO Biological Process and KEGG with clusterProfiler) and over the full ranked list (GSEA). One tumour sample clusters with the normals on PCA; it was flagged and kept, since the paired design turns it into a loss of power rather than a source of false positives.

  • Notebook: CRC_RNAseq_GSE50760_analysis.ipynb (R kernel) — the whole RNA-seq analysis, plus sections 17–18 showing the docking and MD figures beside the PyMOL code that drew them
  • Scripts: 01_RNAseq/scripts/00–08 (R), run in order
  • Record: ANALYSIS_RECORD.md — decisions, figure log and limitations L1–L33

2 · Candidate panel and target selection — 01_RNAseq/

A ten-gene panel (KRT80, ETV4, CDH3, STC2, CGREF1 up; CA4, MS4A12, CLCA4, TMIGD1, CA7 down) was locked and not revised afterwards. All ten were confirmed protein-coding against NCBI and Ensembl, and all ten reproduced in an independent 34-patient cohort (GSE292858). CA7 was chosen for structural work because it combines a strong, reproducible expression change with unusually good structural resources. It is down-regulated in tumour, so it is treated as a structural and biomarker subject — not as a drug target for colorectal cancer.

3 · Zinc-site docking — 02_Docking/

The receptor was PDB 8Q3U (CA7, 1.10 Å) with the co-crystallised ligand, glycerol and all waters removed and the catalytic Zn²⁺ kept (4,219 atoms). Before screening, the crystallographic ligand J8I was re-docked as a control. Plain AutoDock Vina failed; the diagnosis showed the correct pose was being generated but mis-ranked, so more search could not fix it. A controlled 2 × 2 experiment (Vina vs AutoDock4Zn × neutral vs anionic sulfonamide) showed that only the combination recovered the zinc-binding head group.

Twenty ZINC20 compounds were then screened under identical settings (Vina 1.2.7 with --scoring ad4, exhaustiveness 100, seed 5678, 20 modes): eleven primary sulfonamides, six secondary sulfonamides as a hard negative control, and three compounds with no zinc-binding group. Zinc-coordination geometry separated the classes perfectly; the docking score did not — a secondary sulfonamide that cannot coordinate zinc at all scored second best. Chlorthalidone was selected on mechanism and geometry, with score consulted last.

  • Record: DOCKING_RECORD.md — every run, including the failed ones and the Dockey ligand-substitution incident
  • Scripts: 02_Docking/scripts/01–26, numbered in the order they were run
  • Manifest of outputs: results/FINAL_OUTPUTS.md

4 · Molecular dynamics — 03_Molecular_Dynamics/

The docked pose was carried into MD unchanged (verified to 0.000000 Å) and built in CHARMM-GUI with CHARMM36m, CGenFF, TIP3P and 0.15 M NaCl. It was minimised, equilibrated (125 ps NVT, 250 ps NPT with restraints) and run for 6 ns of unrestrained NPT production at 310 K in GROMACS 2026.3 on a GPU. The complex stayed intact, the protein fold was stable and the zinc coordination held throughout.

Two build failures are documented rather than hidden: a ligand atom-name mismatch between CHARMM-GUI modules, and a ligand that came back neutral because the MOL2 format cannot carry a formal charge — caught by auditing the generated topology, and rebuilt from SDF/SMILES.


Selected figures

Click any figure to open it at full size.

RNA-seq and enrichment

PCA
PCA — PC1 (59 %) separates tumour from normal; one tumour sample (AMC_5) sits with the normals and was flagged and kept.
Volcano plot
Differential expression — 2,854 genes at FDR < 0.05 and |log₂FC| ≥ 1 (894 up, 1,960 down).
DEG heatmap
Top DEG heatmap — the strongest up- and down-regulated genes across all twelve samples.
Paired expression
Paired expression — each line joins one patient's tumour to their own matched normal, for the ten panel genes; AMC_5 is the flat outlier.
GO BP up
GO Biological Process, up in tumour — nuclear division, chromosome segregation and DNA replication.
GO BP down
GO Biological Process, down in tumour — loss of differentiated-epithelium and transport functions.
GSEA
GSEA — ribosome biogenesis and DNA replication activated; immune/lymphoid programmes suppressed (all FDR < 0.05).
KEGG down
KEGG, down in tumour — receptor signalling, immune, bile-secretion and drug-metabolism pathways.
Independent validation
Independent validation — the locked panel reproduced 10/10 in the 34-patient GSE292858 cohort.
Pathway-gene network
Pathway–gene network — which genes drive each enriched programme; CA7 sits in the nitrogen-metabolism module.

Docking

J8I redocking
Redocking control — crystal vs AutoDock4Zn pose, no superposition. Zinc anchor reproduced, tail not: a partial validation.
Score versus geometry
The central docking result — classes separate on Zn–N geometry, not on score.
Chlorthalidone pose
Selected pose — chlorthalidone coordinating the catalytic zinc, H-bonded to Thr199.
Active-site contacts
Active-site contacts — measured distances of the docked chlorthalidone pose.

Molecular dynamics (6 ns)

Zinc coordination over 6 ns
Zinc shell — Zn–N 2.214 ± 0.065 Å; four-coordinate in 61/61 frames.
RMSD
RMSD — protein backbone 0.94 Å and ligand 1.55 Å on average.
RMSF
RMSF — mean Cα fluctuation 0.63 Å; the largest (1.89 Å) at residue 262.
SASA
SASA — 87.9 ± 1.6 % of the ligand surface buried; interface 320.4 ± 6.4 Ų.
Stability summary
Stability summary — Zn–N distance, Thr199 contact, radius of gyration and zinc coordination number over 6 ns.
Final pose
Final frame — chlorthalidone at 6 ns, still coordinating the zinc.
Start vs end
Start vs end — the 0 ns and 6 ns poses in the same view.
Ligand in the pocket
Pocket surface — the ligand enclosed in the active site at 6 ns.

Repository layout

.
├── README.md                     this file
├── requirements.txt              Python packages (docking, MD analysis, report)
├── docs/                         the overview figure and the script that draws it
├── 01_RNAseq/                    stage 1–2: RNA-seq, enrichment, validation, target selection
│   ├── CRC_RNAseq_GSE50760_analysis.ipynb
│   ├── ANALYSIS_RECORD.md
│   ├── data_raw/                 GSE50760 counts + GSE292858 validation data
│   ├── scripts/                  00–08 (R), 09–10 (Python helpers)
│   ├── figures/                  Figures 1–20
│   └── results/                  DEG tables, enrichment, validation, target evidence
├── 02_Docking/                   stage 3: receptor prep, redocking, AD4Zn screen
│   ├── DOCKING_RECORD.md
│   ├── inputs/                   hand-prepared AutoDockTools files + Dockey run
│   ├── structures/               8Q3U and the cleaned receptor
│   ├── scripts/                  01–26 (Python, PyMOL, batch)
│   ├── tools/                    README only — third-party tools are not redistributed
│   └── results/                  redocking, screen, final candidate, figures D2–D9
├── 03_Molecular_Dynamics/        stage 4: CHARMM-GUI build, GROMACS run, analysis
│   ├── MD_FINAL_REPORT.md, MD_RUN_LOG.md, CHARMMGUI_RECORD.md
│   ├── charmm-gui_8707215842_ligand/   ligand-parameterisation job (provenance)
│   ├── charmm-gui_8707261082_system/   system-building job (provenance)
│   ├── md_run/                   GROMACS inputs, logs, energies, 6 × 1 ns trajectory
│   ├── scripts/                  01–08
│   └── results/                  per-frame CSVs, Figures M1–M8, final-pose structures
└── 04_Report/                    the written report (PDF + editable DOCX) and its build

Reproducing the analysis

Scripts resolve their paths relative to the repository, so they can be run from any working directory. PyMOL .pml scripts are the exception: run them from inside their stage folder (for example cd 02_Docking then pymol -cq scripts/21_pymol_D2_redocking.pml).

Software used

Stage Software
RNA-seq R 4.6.1, Bioconductor 3.23 — DESeq2, apeglm, clusterProfiler, org.Hs.eg.db (01_RNAseq/scripts/00_install_packages.R)
Docking AutoDock Vina 1.2.7, AutoGrid 4.2.6, AutoDock4Zn, MGLTools 1.5.7, RDKit 2026.03, Meeko 0.7.1, PyMOL 3.1
MD CHARMM-GUI (web), CHARMM36m + CGenFF, GROMACS 2026.3 (CUDA build via 03_Molecular_Dynamics/scripts/install_cuda_gromacs.sh), MDAnalysis 2.10, mdtraj 1.11
Report Python 3.11, Chrome (headless) for PDF, python-docx
pip install -r requirements.txt

Run order

  1. RNA-seq — Rscript 01_RNAseq/scripts/00_install_packages.R, then scripts 01 → 08 in order (script 04 needs internet access for the KEGG REST API). Or open the notebook from inside 01_RNAseq/.
  2. Docking — 02_Docking/scripts/01 → 26 in numbered order. Vina and the AutoDock4Zn tools must be installed first; see 02_Docking/tools/README.md.
  3. MD — the system was built in CHARMM-GUI (jobs 8707215842 and 8707261082, both archived here). Production runs under WSL, from inside 03_Molecular_Dynamics/md_run/: GMX=~/miniforge3/envs/md/bin/gmx CNTMAX=6 bash run_md.sh all. The analysis scripts 03_Molecular_Dynamics/scripts/05–08 need only the trajectories included here.
  4. Report — python 04_Report/build/render.py, print 04_Report/build/report.html to PDF, then python 04_Report/build/make_docx.py.

Not in this repository

Left out to keep the repository small, all either regenerable or third-party:

Excluded How to get it back
01_RNAseq/data_processed/*.rds (≈ 24 MB) written by running scripts 01–08
AutoGrid .map files (≈ 44 MB) regenerate from the saved .gpf — see 02_Docking/tools/README.md
Vina / AutoGrid binaries, AD4Zn.dat, AutoDock4Zn helper scripts official sources in 02_Docking/tools/README.md
GROMACS .tpr, .cpt, .trr, intermediate .gro, equilibration .xtc gmx grompp / run_md.sh from the included inputs and .mdp files
CHARMM force-field folders (toppar/) inside the CHARMM-GUI jobs distributed by CHARMM-GUI; the GROMACS-format copy used for the run is in md_run/toppar/

What this work does and does not show

This is computational structural evidence. Docking scores are scoring-function values, not binding affinities; a stable 6 ns trajectory shows the modelled complex is stable, not that chlorthalidone binds or inhibits CA7. Chlorthalidone is an established carbonic-anhydrase inhibitor, so selecting it is a coherence check on the protocol rather than a discovery. CA7 is down-regulated in colorectal cancer, and nothing here supports inhibiting it as a therapy.

The main limitations are stated alongside the results rather than buried: the redocking validation is partial; the receptor is one rigid crystal conformation with its waters removed; the zinc is a non-bonded ion in MD, so a −1 ligand beside a +2 point charge is attracted partly for electrostatic reasons alone; CGenFF parameterised the deprotonated sulfonamide by analogy to a sulfoximine (penalty 156.5); and there is a single 6 ns trajectory. The full list is in 01_RNAseq/ANALYSIS_RECORD.md (L1–L33) and 02_Docking/DOCKING_RECORD.md (D-L1–D-L19).

Data sources

  • GSE50760 — paired colorectal tumour / normal RNA-seq (NCBI GEO)
  • GSE292858 — independent colorectal cancer cohort used for validation (NCBI GEO)
  • PDB 8Q3U — human CA7 with the inhibitor J8I, 1.10 Å (RCSB PDB)
  • UniProt P43166 — CA7 sequence and zinc-site annotation
  • ZINC20 — the twenty screened compounds

Author

Hossam Mahmoud El-Basiouny — all analyses, figures, records and the report.

Acknowledgements

Carried out as a Biotechnology Summer Project under the supervision of Prof. Abdo El-Fiky and Prof. Ahmed Adel, with the support of the teaching assistants Dr. Omnia Gaafer and Dr. Ibrahim M. Ibrahim.

Releases

Packages

Contributors

Languages