---
title: Interpreting GraphSol outputs
description: >-
  Interpret GraphSol protein solubility probabilities, confidence scores, and
  output artifacts.
---

You get back a solubility probability (0–1) and a binary call for each sequence, alongside summary artifacts. That probability drives your next step: a high value marks a protein worth prioritizing for expression, while a low one flags a likely-insoluble candidate to redesign or re-construct before you invest in the wet-lab work.

## Interpreting GraphSol outputs

Each completed job returns per-sequence predictions and summary artifacts.

## Output fields

- `id`: Sequence identifier from your input.
- `length`: Sequence length in residues.
- `probability`: Solubility probability in the range `[0, 1]`.
- `call`: Binary label using the current threshold (default 0.5).
- `warnings`: Optional messages (e.g., non-standard residues removed).

## Artifacts

- `results.csv`: Tabular predictions with all fields.
- `results.json`: Same content as CSV for programmatic use.
- `summary.json`: Run-level metadata including input counts and threshold used.

## Interpreting scores

- `probability >= 0.7`: High-confidence soluble candidates.
- `0.4–0.7`: Consider further validation or sequence optimization.
- `< 0.4`: Likely insoluble; prioritize for refactoring or alternative
  constructs.

## Adjusting thresholds

If your downstream pipeline prefers higher precision, rerun with a higher
threshold (e.g., `0.6–0.7`). For higher recall, lower to `0.3–0.4`. The CSV is
reusable—no need to resubmit sequences when only thresholding changes.

## Canvas Visualization

When a GraphSol job completes, a result card appears on the canvas.

- **Card size**: 350x300px
- **Border color**: Orange (#FF9800)
- **Visualization**: Solubility prediction with graph-based feature visualization

### Tier 2 Metrics

- Solubility score
- Confidence percentage
- Graph features (nodes, edges)
- Important residues highlighted by attention weights

## Read next