Case Studies
TARGET DISCOVERY & ID
ENTPRISE: An algorithm for predicting human disease-associated amino acid substitutions from sequence entropy and predicted protein structures
A knowledge based approach for predicting gene-disease associations
Fast Procedure for Reconstruction of Full-Atom Protein Models from Reduced Representations
TASSER_WT A Protein Structure Prediction Algorithm with Accurate Predicted Contact Restraints for Difficult Protein Targets
Template-based protein structure modeling using TASSERVMT
Fr-TM-align - a new protein structural alignment method based on fragment alignments and the TM-score
iAlign - a method for the structural comparison of protein-protein interfaces
Insights into disease associated mutations in the human proteome through protein structural analysis
Comprehensive prediction of drug-protein interactions and side effects for the human proteome
A threading-based method for the prediction of DNA-binding proteins with application to human genome
EFICAz2 - Enzyme Function Inference by a Combined Approach Enhanced by Machine Learning
APoc - large-scale identification of similar protein pockets
TARGET Discovery and ID
SUMMARY OF RESULTS- The purpose of this case study was to experimentally validate PanXome’s ENTPRISE software. ENTPRISE is a boosted tree regression machine-learning approach to predict human disease-associated amino acid variations by utilizing a comprehensive combination of protein sequence and structure features. To validate ENTPRISE, it was compared to other current state of the art methods. ENTPRISE outperforms all other methods significantly with a MCC of 0.645 compared to the next best method FATHMM with a MCC of 0.538. Except for a sensitivity of 77%, ENTPRISE has the best values for all other measures. ENTPRISE has the best true positive rate at all false positive rate levels. At a 10% false positive rate, ENTPRISE has a true positive rate of 54%; whereas the next best method, MUTATIONASSESSOR has a true positive rate of 42%.
LEAD IDENTIFICATION
SUMMARY OF RESULTS- Six disparate proteins of medical interest were used to validate PanXome's software. The top 1% of binding ligands predicted for these proteins by PanXome software were tested using thermal shift assays. The average successful hit rate of the predictions for the test set was 24%, which is dramatically higher than conventional hit rates, which are about 1%. About 50 ligands per protein gave interpretable thermal shift curves. Of these 300 ligands, 47 ligands or 16% had binding affinities with µM or better and 10 or 3% of these had binding affinities of nM or better.
DRUG DE-RISKING
SUMMARY OF RESULTS- Six drugs that have recently failed safety trials were run through PanXome software. The side effects, kill indices, number of off target interactions and a list of all other proteins that the drug interacted with were predicted for all six of the drugs. PanXome's software was able to accurately predict the side effects for the six high profile safety failures and therefore correctly predicted the reason for their withdrawals. One drug in particular, Vioxx, was investigated further and all side effects and implicated proteins that caused its FDA failure were predicted. This validates PanXome's software and shows its use by allowing companies to understand the potential side effects that will cause drugs to fail safety trials before entering clinical trials.
DRUG REPURPOSING
SUMMARY OF RESULTS- Approximately 6,000 drugs were screened against a target library of 33,000 proteins. The median number of protein targets per drug was 38, showing that most drugs likely have multiple targets. Similarly, the median number of drug interactions per protein is 55, showing that most proteins likely interact with many drugs. This demonstrates that each drug has an average of 38 protein targets in the body and therefore many more potential uses. PanXome can be used to identify new targets so existing drugs in the market can be repurposed to new diseases. Fifteen drugs that recently failed efficacy trails were investigated further and of these, PanXome software was able to predict alternative protein targets for eleven of the drugs.