iSEE is a computational predictor of binding affinity changes upon single point mutations (∆∆G) [1]. It is based on a random forest model that makes use of interface Structure, Evolution and Energy-based features.

This repository provides the iSEE predictor that you can use to do ∆∆G prediction and to identify the effects of mutations on your protein complexes. Besides, we provide all features data used in the iSEE project [1].

The HADDOCK-refined models of both wild type and mutant complexes as well as the PSSM files used to construct the features can be found in the SBGrid data repository (doi:10.15785/SBGRID/520).

Please cite the article [1] if you use iSEE.

1. Download and install R. After installing, type command R or RScript in your terminal, they should be available now.
2. Install R packages randomForest and caret

# Enter R enviroment
R

# Install R packages
> install.packages(c("caret", "randomForest"))

# Quit R environment
> q()

To use the iSEE ∆∆G predictor at the command line type:

Rscript run_iSEE.R  <features data file>

• iSEE.model is the iSEE ∆∆G predictor, which is a binary file and executed by the script run_iSEE.R. Note the two files must be in the same directory.
• Users need to provide the <features data file> with rigorous format as shown in the example file features_examples.tsv. Headers must be present and a tab should be used as column separator.

Rscript run_iSEE.R features_examples.tsv

The directory isee_features_data contains all features data used in this work[1].

Features based on the top 1 HADDOCK refined model:

• features_training_dataset_top1.tsv: Features used as training dataset. The iSEE.model we provide here is trained on these data.
• features_NM_test_dataset_top1.tsv: The independent NM test dataset of Benedix et al. Nature Methods 2009.
• features_S487_test_dataset_top1.tsv: The independent SKEMPI2 S487 test dataset.
• features_MDM2-p53_test_dataset_top1.tsv: The independent MDM2-p53 test dataset.
• features_MDM2-p53_full_mutations_top1.tsv: Features used for MDM2-p53 full mutations.

Average features based on the top 4 HADDOCK refined models:

• features_training_dataset_top4.tsv
• features_NM_test_dataset_top4.tsv
• features_S487_test_dataset_top4.tsv
• features_MDM2-p53_test_dataset_top4.tsv
• features_MDM2-p53_full_mutations_top4.tsv

1. C. Geng, A. Vangone, G.E. Folkers, L.C. Xue and A.M.J.J. Bonvin. iSEE: Interface Structure, Evolution and Energy-based machine learning predictor of binding affinity changes upon mutations. Proteins: Struc. Funct. & Bioinformatics Advanced Online Publication (2018).