Skip to content
RCreddit.com·

Why protein-structure confidence scores cannot substitute for a biological assay

AI summary

Protein-structure confidence scores cannot fully substitute for biological assays because a FASTA sequence alone does not fully specify a prediction job. Different inputs and validation are needed for single chains, protein assemblies, and protein-ligand complexes. Database versions, alignments, templates, and sampling protocols also influence the tested outcome. Confidence scores, whether local, relative domain placement, or interface confidence, address different aspects of predicted geometry but do not measure catalytic activity or binding. The key is to evaluate which biological claim the output supports and what evidence is needed before using a model-selected structure for functional claims.

Time & source
Published
09/07, 15:36 UTC+0
Ingested
09/08, 00:00 UTC+0
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

Article · full text via RSS

Disclosure: I run rewire.it, where I published the article linked below.

A FASTA sequence does not fully specify a protein-structure prediction job. A single chain, a protein assembly and a protein–ligand complex require different inputs and validation. Database versions, alignments, templates and the sampling protocol also change what was actually tested.

Confidence scores need the same distinction. Local confidence, relative domain placement and interface confidence concern different aspects of the predicted geometry. None is a measurement of catalytic activity or binding in a particular assay. Drawing more diffusion samples can explore candidate structures, but those samples are not automatically a thermodynamic ensemble.

I wrote a guide to those choices, including what to record before comparing models or interpreting their scores: https://rewire.it/blog/a-fasta-file-is-not-a-specification/

The useful evaluation question is which biological claim the output supports. What evidence would you require before carrying a model-selected structure into a downstream functional claim?

Source·reddit.com·Full text via RSS