FAIDR-Fit

Explore the molecular features and function of intrinsically disordered regions (IDRs) across the human proteome in your browseri

Enter a group of proteins sharing an associated function or characteristici

Or choose a predefined group

Or randomize a number of IDRs

Results

Number of molecular featuresi

0 molecular features selected

Recall: — Precision (Enrichment): —

enriched (+z) depleted (−z) i

About FAIDR-Fit

FAIDR is a statistical model for studying intrinsically disordered regions (IDRs), introduced in Zarin et al., eLife (2021) (opens in new tab) and applied to the human proteome in Pritišanac et al., PNAS (2026) (opens in new tab). This web app provides an online version of it.

FAIDR-Fit implements a slightly simplified version of the classifier described in the papers above. It was written with help from AI (Claude Opus 4.8 and Claude Sonnet 5) and refined through significant trial and error and manual checking. Read the report (PDF) (opens in new tab) for more details. The web app, code, and report are provided under CC BY-NC 4.0 (opens in new tab), as is and without warranty of any kind.

If you use FAIDR-Fit in your research, please cite this tool and the Pritišanac et al. paper above. See the README (opens in new tab) on GitHub for citation details.

How FAIDR works

FAIDR stands for Feature Analysis of IDRs. Given an initial group of proteins, it extracts molecular features of their IDRs, specifically how conserved each feature is across evolution, and builds a model that identifies which of 144 molecular features are most relevant for predicting group membership.

Once the model is built, a prediction for every single IDR in the human proteome is available. This allows exploration of proteins sharing similar conserved molecular features in their IDRs.

Contact

guillaume.koenig@mail.utoronto.ca · LinkedIn (opens in new tab)

Suggestions and feedback are welcome. You can also open an issue (opens in new tab) on the project's GitHub page.

Defining a function group

Proteins should be entered as a column (one per line) as uppercase UniProt accession IDs. The easiest way is to copy and paste from a spreadsheet column. Input is lenient: non-matches and duplicates are automatically discarded. All the IDRs of a protein are considered. (Unless using specific IDR ids, as found in the list of predictions.)

Groups of proteins are commonly retrieved from experiment or from GO terms.

For example, suppose you are interested in cytoplasmic stress granules, GO:0010494 (opens in new tab) at EBI QuickGO. On the annotation page, filter by Homo sapiens, and export all annotations as TSV (Tab-Separated Values). In that spreadsheet, copy the column "GENE PRODUCT ID" of UniProt accession IDs into the main web app text area.

Alternatively, you can pick from a predefined list of GO-term-based groups (from Pritišanac et al., PNAS 2026), or randomize a group to gauge baseline performance. A minimum of 100 proteins is advised for adequate performance, as smaller groups tend to show enrichment similar to a random group of the same size.

Selecting a number of molecular features

You can choose from 1 to 30 molecular features in the model (not every number in that range may be available). The recall and precision scores tend to be better with more features; however, a model with fewer features is often more desirable when scores change only marginally.

To gauge how predictive your input group actually is, compare its enrichment to models built from the same number of IDRs picked at random (see the Randomize button). A meaningful group typically shows enrichment several times higher than a random one of the same size.

Predictions of the model for all IDRs in the human proteome are available for download as a TSV file (Tab-Separated Values). In that spreadsheet, column 1 is the IDR ID, column 2 the UniProt entry name, column 3 whether in the initial group (0/1), column 4 whether predicted to belong to the group (0/1), column 5 the prediction probability (with threshold given in metadata at the top).

Reading the bar graph

The bar graph shows the strength (z-stat) of conserved molecular features which are best at predicting whether an IDR belongs to the input group and its associated function.

The emphasis on conserved is key. It's not simply the molecular features of IDRs that are considered, but how well they are conserved over evolutionary time.

Enriched (red bars) means the feature is conserved and predictive of membership in the input group; depleted (blue bars) means the feature is conserved and predictive against membership in the input group.