Specialized protein prediction database

SUMILE-TM Atlas

Amino-acid-level protein–protein interface-site predictions for reviewed UniProt alpha-helical transmembrane proteins.

Current release2026_08_30 · v1SUMILE-TM headline ensemble

Atlas query

Filter-based search

Combine protein annotations, taxonomy, sequence properties, and SUMILE-TM prediction attributes in one query.

1 condition · maximum 8

AND or OR applies to the complete condition set. Exclude applies NOT to one condition. At least one included condition other than record type is required; annotation groups may overlap.

Atlas coverage

80,619
protein records
70,247
unique sequences
33,722,948
residue predictions
80,062
exact-length records
557
windowed records
189
length-OOD records

Prediction method

SUMILE-TM combines six selected residue learners spanning three protein language models. The canonical atlas value is the raw ensemble probability for each residue.

Binary threshold
0.554184
Threshold source
SI30 validation positive-F1
Task
Partner-independent PPI interface-site prediction

Data access

Protein records, residue-range JSON, per-protein TSV files, and small research batches are available. Whole-atlas download remains pending while the access policy is finalized.

  • Up to 10 accessions per batch request
  • Up to 50,000 residues per batch response
  • No unfiltered accession listing