Function Prediction ServerKiharaLab
Four predictors plus GO2Sum summaries

Protein Function Prediction

Automated GO term prediction using distantly related sequences, phylogenetic distance weighting, iterative similarity search, and domain embeddings, with GO2Sum generating UniProt-like functional summaries from predicted GO terms.

Start PredictingView Documentation

PFP

Uses distantly related sequences and contextual associations of GO terms with PSI-BLAST E-values up to 125.

Phylo-PFP

Reranks sequences by phylogenetic distance using MUSCLE and PHYLIP for improved accuracy.

ESG

Iterative PSI-BLAST with probability scores from multiple-level neighbors in a similarity graph.

Domain-PFP

Predicts function through self-supervised learning of function-aware domain embedding representations.

GO2Sum

Creates UniProt-like functional summaries for all predictions from the predicted GO terms.

Features

Comprehensive Function Prediction

Our tools provide detailed GO term predictions with probability scores across all three ontology categories.

Distantly Related Sequences

PFP retrieves sequences with E-values up to 125, far beyond conventional thresholds, for broader coverage.

Phylogenetic Distance Reranking

Phylo-PFP constructs phylogenetic trees using MUSCLE and PHYLIP to rerank sequences by evolutionary distance.

Contextual Associations

GO term co-occurrence statistics improve predictions by considering functional relationships between terms.

Iterative Search

ESG recursively performs PSI-BLAST to find multi-level neighbors in protein similarity graphs.

Multiple Sequences

Submit one or more protein sequences in FASTA format for batch processing.

Confidence Levels

Results are classified into very high, high, moderate confidence levels based on normalized scores.

Tools

Choose Your Method

Select the predictor or summary tool that best fits your workflow, from fast baseline GO prediction to UniProt-like summary generation.

PFP
Extracts GO annotations from distantly similar sequences (E-value up to 125) and applies contextual associations of GO terms observed in the annotation database.
  • PSI-BLAST with E-value up to 125
  • Contextual GO term associations
  • Ranked best in CASP7 function prediction
Use PFP
Phylo-PFP
Improved PFP that reranks sequences by phylogenetic distance using MUSCLE alignment and PHYLIP neighbor-joining trees with PROTDIST.
  • Evolutionary distance-normalized scoring
  • MUSCLE MSA + PHYLIP tree construction
  • Outperformed top CAFA2 methods
Use Phylo-PFP
ESG
Recursive PSI-BLAST search that assigns probability scores based on relative similarity to multiple-level neighbors in a protein similarity graph.
  • Iterative PSI-BLAST searches
  • Multi-level neighbor weighting
  • Ranked 4th in CAFA for MF predictions
Use ESG
Domain-PFP
A self-supervised method that learns functionally consistent representations for domains through domain-GO co-occurrences and associations.
  • Self-supervised domain embeddings
  • Function-aware domain representations
  • Superior to large-scale language models
Use Domain-PFP
GO2Sum
A fine-tuned T5-based summarization tool that turns predicted GO terms from all prediction tools into UniProt-like function, pathway, and subunit text.
  • GO-term-to-paragraph translation
  • Fine-tuned T5 encoder-decoder model
  • Summarizes predictions from all tools
Learn GO2Sum

Ready to Predict Protein Functions?

Upload your FASTA sequence and get GO term predictions in minutes. No registration required.

Start Predicting NowRead the Docs