---
title: Post-export playbook
description: >-
  Verification, external analysis integration, and data archival after exporting
  results from the platform.
tags:
  - nextsteps
  - post-export-playbook
---

Once you have exported results from the platform -- CSV tables, SDF files, PDB
structures, JSON reports, or PDF passports -- verify those outputs, feed them
into external analysis tools, and archive everything for reproducibility.

> A project export (a ZIP) includes `project_metadata.json` and, for every job,
> a `jobs/<jobId>_metadata.json` sidecar. The sidecar records the tool name,
> status, parameters, result summary, storage references, and created, updated
> and completed timestamps. It does not carry file checksums, so record your own
> at download time (see below).

---

## Verification after export

### File integrity checks

Before loading exported data into any downstream tool, confirm that files
arrived intact.

1. **Row and column counts** -- Open CSV/TSV exports and verify that the
   row count matches the count displayed on the platform results panel. Off-by-
   one differences usually mean a header row was included or excluded.
2. **Structure file validation** -- For PDB and SDF files, run a quick
   sanity check with Open Babel (`obabel -ipdb input.pdb -osdf -O /dev/null`)
   or RDKit (`Chem.MolFromMolFile(path)`) to catch truncation or encoding
   errors.
3. **Checksums** -- Record a SHA-256 for each file when you download it, and
   store the list next to the exports. Re-run it before any later use to catch
   corruption or accidental edits:
   ```bash
   sha256sum exported_file.csv >> SHA256SUMS
   sha256sum -c SHA256SUMS
   ```
4. **JSON shape checks** -- If you consume JSON exports programmatically,
   assert the fields your pipeline depends on before ingesting, rather than
   assuming a fixed schema.

### Common integrity issues

| Symptom | Likely cause | Fix |
|---------|-------------|-----|
| Truncated CSV | Download interrupted | Re-export from the platform job page |
| Garbled Unicode in compound names | Encoding mismatch | Ensure your tool reads UTF-8 |
| SDF with no 3D coordinates | The tool emitted 2D or SMILES-only output | Generate conformers (for example with RDKit) before 3D work |
| Missing columns in CSV | The export covered a summary, not the full result | Export the full result, or ask Ivy to read the full result in chat or in the chat's [code workspace](/docs/sandbox) |

---

## External analysis tools integration

### GraphPad Prism

Prism is widely used for dose-response curve fitting and statistical
analysis in pharmacology.

- **Import format**: Export dose-response data from the platform as CSV. Prism
  expects columns for concentration, response, and replicate identifiers.
- **Curve fitting**: Use the `log(inhibitor) vs. response -- Variable slope`
  model to derive IC50, Hill slope, and confidence intervals.
- **Batch import**: For multi-compound screens, export a single CSV with a
  `compound_id` column and use Prism's "Group" feature to fit curves per
  compound.

### TIBCO Spotfire

Spotfire excels at interactive visualization of large compound libraries
and SAR (structure-activity relationship) data.

- **SMILES rendering**: Import SDF or CSV with a SMILES column. Spotfire's
  chemistry panel renders 2D depictions automatically.
- **Trellis plots**: Use the platform's exported potency, selectivity, and ADMET
  columns to build multi-panel trellis plots for SAR exploration.
- **Filtering**: Apply Spotfire data functions to flag compounds that fail
  Lipinski's Rule of Five or have PAINS alerts, cross-referencing the platform's
  ADMET predictions.

### MATLAB and Python

For custom analysis beyond what the platform provides:

- **MATLAB**: Load CSV exports with `readtable`. Use the Bioinformatics
  Toolbox for sequence analysis or the Curve Fitting Toolbox for custom
  pharmacokinetic models.
- **Python (pandas/RDKit)**: Read CSV into a DataFrame. Use RDKit to
  recompute descriptors, generate fingerprints, or run substructure
  searches on exported SMILES.

  ```python
  import pandas as pd
  from rdkit import Chem
  from rdkit.Chem import Descriptors

  df = pd.read_csv("ivy_export.csv")
  df["MolWt"] = df["smiles"].apply(
      lambda s: Descriptors.MolWt(Chem.MolFromSmiles(s))
  )
  ```

### PyMOL and ChimeraX

For structural biology exports (PDB, mmCIF):

- **PyMOL**: Load exported PDB files directly. Use `spectrum b, blue_red`
  to color by pLDDT confidence if the B-factor column contains the platform's
  confidence scores.
- **ChimeraX**: Load exported PDB or mmCIF files and color by the B-factor
  column (for example `color bfactor palette alphafold`) to show per-residue
  confidence from a structure predictor.

---

## Data archival best practices

### Folder structure

Adopt a consistent directory hierarchy for every project:

```
project_name/
  raw_exports/          # Unmodified .bloom* exports
  processed/            # Post-analysis outputs (Prism fits, Spotfire DXPs)
  figures/              # Publication-ready figures
  metadata/             # _metadata.json, parameter logs
  notebooks/            # Jupyter/R notebooks with analysis code
```

### Version control for analysis code

Track all post-export scripts in Git. Include:

- The exact platform job IDs and export timestamps in commit messages.
- A `requirements.txt` or `environment.yml` so collaborators can reproduce
  your environment.
- The `SHA256SUMS` file you recorded at download time, and a script that re-checks it.

### Long-term storage

- **Institutional repositories**: Deposit finalized datasets in Zenodo,
  Figshare, or your institution's data repository with a DOI.
- **Cloud backup**: Sync `raw_exports/` to S3, Google Cloud Storage, or
  Azure Blob. The platform's export filenames include job IDs, making deduplication
  straightforward.
- **Retention policy**: Keep raw exports for a minimum of 10 years for
  regulatory submissions. Processed data should be retained at least as
  long as the associated publications.

---

## Sharing results with collaborators

### Direct sharing from the platform

- **Share links**: Generate a read-only share link from any job page.
  Recipients can view results, 3D structures, and charts without a platform
  account.
- **Project-level sharing**: Add collaborators to a platform project to grant
  access to all jobs, exports, and chat history within that project.

### Benchling

Benchling is widely adopted in biopharma for molecular biology workflows
and inventory management.

- **Sequence registration**: Export protein sequences from the platform in FASTA
  format and register them as Benchling entities. Attach the platform job URL
  as a linked reference.
- **Notebook entries**: Create a Benchling notebook entry for each
  computational campaign. Embed exported figures and link to the platform
  project share URL.
- **Custom schemas**: If your Benchling instance has custom entity schemas
  for compounds, map the platform's SMILES, InChIKey, and ADMET predictions to
  schema fields during import.

### Other electronic lab notebooks

- **Signals Notebook (Revvity Signals)**: Import CSV exports and attach SDF
  files to experimental entries.
- **BIOVIA Notebook**: Use the REST API to programmatically push platform
  exports into notebook sections.
- **RSpace**: Drag-and-drop platform PDF reports into RSpace entries.

---

## Quality checks on exported data

### Statistical sanity checks

Before using exported data for decision-making:

1. **Distribution inspection** -- Plot histograms of key properties
   (docking scores, ADMET predictions, pLDDT values). Unexpected bimodal
   distributions or outliers may indicate input errors.
2. **Cross-tool consistency** -- If you ran the same compounds through
   multiple platform tools (e.g., ADMETLab3 and ADMET-AI), compare predictions.
   Large disagreements warrant manual inspection.
3. **Known-compound benchmarking** -- Include 2-3 reference compounds with
   known experimental values in every batch. Compare the platform's predictions
   against published IC50, solubility, or permeability data.

### ADMET prediction validation

| Property | Typical benchmark | Source |
|----------|------------------|--------|
| LogP | Measured vs. predicted R^2 &gt; 0.85 | ChEMBL lipophilicity dataset |
| Solubility | Within 0.5 log unit of measured | [AqSolDB](https://doi.org/10.1038/s41597-019-0151-1) |
| hERG inhibition | Binary accuracy &gt; 80% | [Karim et al. 2021](https://doi.org/10.1186/s13321-021-00541-z) |
| CYP450 inhibition | AUC-ROC &gt; 0.80 | PubChem BioAssay |

### Structure quality metrics

For protein structure exports, check:

- **Ramachandran outliers** < 2% (use MolProbity or the platform's built-in
  quality panel).
- **Clashscore** < 10 for ESMFold/ColabFold predictions.
- **pLDDT mean** > 70 for the regions you plan to use in docking.

> Never use a predicted structure with mean pLDDT below 50 for binding
> site analysis. Low-confidence regions should be rebuilt or excluded
> before docking.

---

## Checklist

Use this checklist after every export:

- [ ] File sizes and row counts match expectations
- [ ] Checksums recorded at download and re-verified
- [ ] Structure files parse without errors in RDKit or Open Babel
- [ ] Reference compounds included and predictions are within expected range
- [ ] Raw exports archived unmodified, with the download time recorded alongside them
- [ ] Analysis code committed to version control
- [ ] Share links or ELN entries created for collaborators
