maxent-ppi

Java executable jar to train and evaluate protein-protein interactions based on ontology annotations using a Maximum Entropy model.

Datasets

Datasets based on CYC2008 v2.0 a S. cerevisiae curated protein complex data set and 1,379 binary interactions observed in at least one out of 14 biochemical conditions Celaj et al., 2017.
Dataset folder:

• protein/gene GO annotaiton files: go2ppi_fbgn.anno and go2ppi_gene_symbol.anno
• protein/gene interaction files: go2ppi_fgbn.ppi and go2ppi_gene_symbol.ppi (first 500 positive set, last 500 negative set)
• Gene Ontology OBO file: gene_ontology_edit_01_12_11.obo
• Yeast 2017: yeast_ppis_ORF_1379.txt and yeast_ppis_ORF_1379_IDs_624.txt (positive set); lowConf_yeast_ppis1379_IDs618_ppis.txt and lowConf_yeast_ppis1379_IDs618_prots.txt (negative set); ppi_GO_merged.txt (postive and negative set together)

Maximum Entropy package

gis-maxent

• original opennlp-maxent-3.0.0-src.tar.gz package downloaded from the OpenNLP Maximum Entropy Sourforge project
• additional modification to the code of opennlp.maxent.GISTrainer: offer correction constant alternative to maximum length of a feature vector (previous implementation) to median length of feature vector (current implementation) GISTrainer.java#L306-L326

Functionality

Java wrapper functions:

• training:
  java -jar maxent-ppi-wrapper.jar -train -i train.dat -o model.out
• evaluation:
  java -jar maxent-ppi-wrapper.jar -eval -i test.dat -m model.out -o test.out
• evaluation with raw scores:
  java -jar maxent-ppi-wrapper.jar -evalScore -i test.dat -m model.out -o test.out
• export weights:
  java -jar maxent-ppi-wrapper.jar -weight -i model.out -o modelWeights.out

Reference

Open access online publication:
Armean,I.M. et al. (2018) Co-complex protein membership evaluation using Maximum Entropy on GO ontology and InterPro annotation. Bioinformatics, 34, 1884–1892. DOI: https://doi.org/10.1093/bioinformatics/btx803