Supervised Learning of Phosphopeptide Sequence Constraints Enables Global Prediction of SH2 Domain Binding

In plain English

The preprint asks a broad first question before predicting a specific interaction: does the amino-acid sequence around a phosphotyrosine look compatible with the shared recognition rules of the SH2 family at all? The authors train a classifier from diverse binding datasets and use it to annotate more than 45,000 human phosphotyrosine sites, prioritize candidates for enrichment experiments, and analyze mutation effects.

How the study worked

A plain-language walk through the work behind the result.

Integrated phosphopeptide-binding data spanning 120 SH2 domains into a supervised sequence classifier.
Applied the classifier to the human phosphoproteome, individual experiments, super-SH2 enrichment, and mutation-effect analyses.

What they found

A relatively small, representative binder set was sufficient for the reported shared-family classification task.
The model produced a global annotation of potential SH2-binding participation across more than 45,000 human phosphotyrosine sites.

Why it matters

A family-level filter could reduce the search space before researchers attempt the harder task of assigning a phosphosite to a particular SH2 domain.

The catch

The work is a preprint and has not completed peer review.
The classifier predicts general SH2-binding potential, not a specific domain–phosphosite interaction.
The abstract does not report full external-validation or subgroup-performance details.

Methods at a glance

Study type: Supervised phosphopeptide-sequence classification
Biological system: Human phosphotyrosine sites and SH2-domain recognition
Organism: Human
Dataset: Integrated phosphopeptide-binding data across 120 SH2 domains and the human phosphoproteome
Sample size: More than 45,000 human phosphotyrosine sites and data spanning 120 SH2 domains
Model architecture: Supervised sequence classifier; architecture details are not reported in the abstract
Training objective: Classify phosphotyrosine-centered sequences as compatible or incompatible with shared SH2 recognition constraints
Code: Code is available at https://github.com/NaegleLab/SpY-C.
Data: The authors link Figshare records at https://doi.org/10.6084/m9.figshare.33256416 and https://doi.org/10.6084/m9.figshare.33249876.

Tasks and metrics

Tasks: phosphoproteome annotation; candidate enrichment; mutation-effect analysis

Metrics: Classification metrics are not reported in the abstract

Baselines: Baseline details are not reported in the abstract

Principal findings

The authors report that a relatively small representative binder set supported family-level classification.
SpY-C annotated potential SH2-binding participation across more than 45,000 human phosphotyrosine sites.

What remains unresolved

External validation: The abstract reports a wide range of validation but does not describe the validation design.

The preprint has not completed peer review.
The output is a family-level participation score rather than a specific SH2-domain interaction prediction.
The abstract does not provide the complete benchmark design or subgroup results.