Interpreting Results

Reading the output

What each panel means and how to act on it.

pLDDT — structure confidence

pLDDT (predicted Local Distance Difference Test) is a per-residue confidence score reported by structure prediction models. foldfunc reports the mean across the full sequence. Use it as a guide to how much to trust the predicted fold, not as a quality score for the protein itself.

90 – 100

Very high confidence

Well-ordered region, likely reflects the true structure.

70 – 90

Confident

Generally reliable. Minor local errors possible.

50 – 70

Low confidence

Treat with caution. May indicate flexibility or prediction uncertainty.

< 50

Very low / disordered

Likely intrinsically disordered. Predicted coordinates are not meaningful.

3D structure viewer

The viewer renders the predicted structure in cartoon representation. The colour scheme runs spectrally from blue (N-terminus) to red (C-terminus) — not by confidence. To get a sense of confidence per-region, correlate the coloured ribbon with the pLDDT score.

You can rotate (click + drag), zoom (scroll), and pan (right-click + drag). The viewer is interactive and runs entirely in the browser — no data is sent at this stage.

Mutation scores

Each scored position displays a wild-type log-probability — how expected that amino acid is at that position according to evolutionary patterns learned from millions of protein sequences. The heatmap colour reflects mutational tolerance:

Green (score near 0) — Evolutionarily expected. Mutations here are likely tolerated.

Amber (score −1 to −3) — Moderately conserved. Mutations may reduce fitness or stability.

Red (score < −3) — Highly conserved. This position is likely critical — mutations are probably deleterious.

The top suggested substitution at each position is also shown — the amino acid the model considers most probable at that site given the surrounding sequence context.

AI interpretation

The AI panel is generated by an LLM and structured into four parts:

Protein family

The LLM's best-guess classification based on sequence patterns and literature context. Treat as a hypothesis, not a definitive annotation.

Structural observations

Key observations about the predicted structure — notable domains, disordered regions, and how confidence distributes across the sequence.

Research questions

Open questions the analysis raises. Useful for framing next experimental steps.

Confidence note

An honest assessment of prediction reliability given sequence length, pLDDT, and available literature. Lower-confidence outputs are flagged explicitly.

AI outputs are probabilistic and may contain errors. They should inform scientific thinking, not replace experimental validation. Do not use for clinical or medical decisions.

Literature snippets

Up to 3 paper titles are shown, selected by relevance to your protein name. These are the same publications used to ground the AI interpretation. Click any title to read the full abstract on PubMed. If no protein name was provided, this panel will be empty.

← How It Works FAQ →