Drug Repositioning Platforms to Predict Drug- Disease Signatures

Transcription

1 Drug Repositioning Platforms to Predict Drug- Disease Signatures

2 Drug Repositioning Screen Typical Experiment screening (mostly Irrational) Experimental Methods Transcriptome matching, cell based models Phenotypic screening Protein-protein interaction assays Drug annotation technologies Gene activity mapping In-vivo animal models Observational studies Experimental medicine (case based studies)

3 Cons of Experiment Screening o Experimentally testing drugs against all targets is extremely expensive, and technically not achievable o Even after finding a hit the rational mechanism of action must still be revealed and tested o Not every drug is available commercially, If available, is expensive from the branded company o Set up an assay from scratch is a daunting task, even after setup, not guarantee FDA drugs are going to work o The financial cost per assay run, and, maintaining the target immobilization may alter binding site properties o The amount of potential druggable targets is exponentially increasing, creating the vast possible drugtarget interactions space explore with expt. is a challenge.

4 Not Drug Repositioning Screen In-Silico Methods Computer Based screening (Rational) Target based Protein binding site based similarity Protein-Ligand Docking Simulations Chemical centric Physico-chemical properties Chemical similarity Shape similarity Other methods Drug disease relationships, Drug regulated gene expression, Side effect profile, Literature Mining of smallmolecules

5 Drug Repositioning Screen Docking = virtual screening gold standard Structure based fit ligands into the real atomic structure of the protein or use homology model P ><SF > < S6/M2 > KcsA 65 ALWWSVETATTVGYGDLYPVTLWGRLVAVVVMVAGITSFGLVTAALATWF VGREQE MthK 364 SLYWTFVTIATVGYGDYSPSTPLGMYFTVTLIVLGIGTFAVAVERLLEFL INREQM KvAP 198 ALWWAVVTATTVGYGDVVPATPIGKVIGIAVMLTGISALTLLIGTVSNMFGKIL----VGEPEP ratkv AFWWAVVSMTTVGYGDMVPTTIGGKIVGSLCAIAGVLTIALPVPVIVSNFNYFYHRETEGEEQA hkv AFWWAVVTMTTVGYGDMHPVTIGGKIVGSLCAIAGVLTIALPVPVIVSNFNYFYHRETEGEEQS hkv AFWWAVVTMTTVGYGDMRPITVGGKIVGSLCAIAGVLTIALPVPVIVSNFNYFYHRETDHEEPA hkv SFWWATITMTTVGYGDIYPKTLLGKIVGGLCCIAGVLVIALPIPIIVNNFSEFYKEQKRQEKAI Docking, MD simulations

6 Drug Repositioning Screen Virtual Screening Performance (existing computer screening methods) Poor correlation of K i values with docking score (2,335 holo protein structures) Poor correlation of IC 50 values with docking score (2,335 holo protein structures) Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article))

7 Challenges to Virtual Screening Inadequate treatment of solvent effect: Solvent water and counter ions are needed to treat proper volume, change shape of the binding site, bridging interaction Conformational change and entropy not included Good affinity prediction not necessarily leads to correct biological binding mode Inadequate treatment of protein (and drug) dynamicity i.e. docking treat protein as rigid All these factors contribute to high false positives and false negatives in virtual screening

8 Drug Repositioning Screen Given low hit rate, Need a New Comprehensive Method? o TMFS method using Proteo-chemometric approach combines ligand and protein centric to predict high hit rate. o TMFS method is combined to systems medicine, patient survival data with gene expressions to predict mechanism of action and make drug-disease relationships for new therapeutics (standalone or combination) 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article) 3. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

9 TMFS Method Train: Match: Fit: Train the drugs -> Ref. ligands, proteins (topology, properties), expt. data Match their topology, properties Fit the data with expt. Data, and integrate it with systems medicine pipeline Streamline: Streamline the hits using the TMFS score 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article) 3. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

10 TMFS Method Algorithm Z score rank ordered viz. top 1 through top 40 target hits for each drug Z Y(, ) k p l 1 i 1 [ f i (, ) p l i 1 f (, )] c l j n 1 X n (, ) CS( OLIC) c l normalized docking score normalized Euclidean distance scores protein pocket, FDA drugs, Ref. native ligands Ligand based descriptor score Ligand shape, physicochemical properties of FDA drugs, Ref. native ligands Ligand topology descriptor score surface properties of FDA drugs, Ref. native ligands Ligand - Protein contact points score CS (OLIC) = S (OLIC-R) - S (OLIC-T) contact point score of FDA drugs contact point score of reference ligands 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article) 3. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

11 TMFS Method Algorithm RePurposeVS Method 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article))

12 RePurposeVS Z-rank score Example: Sutent (Sunitinib) -> predicted alternate targets 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article))

13 Performance of the RepurposeVS o 37% increase in enrichment at the top 1% compared to GLIDE (commercial software) o 48% increase in enrichment at the top 5% compared to GLIDE.

14 Application: Drug Repositioning Alternative Targets of FDA Blockbuster Drugs

15 FDA Blockbuster Alternate Targets Sutent (Sunitinib) is predicted to hit the greatest number of protein targets followed by Alimta. Prograf, Valcote, Concerta, Sifrol, Niaspan, Exelon, Evodart, Sevorane, and Klacid have no hits

16 Repurposing Potential New Disease Classes Anti-neoplastic drugs have the greatest repurposing potential Anti-infective drugs have low repurposing potential Anti-psychotic, anti-bacterial drugs have modest repurposing potential

17 Validation of Predictions To calculate the percent correctly predicted (PCP) targets, PCP i 1 ( nb ji na nx ji ji nb ny ji ji ny nz ji ji ne ji ) x100 where i < # drugs, j < #targets o RpurposeVS predicts drug-target associations with greater than 91% accuracy for the majority of drugs

18 NextGenRepurposeVS: Connected to Molecular Mechanistic Pathways and Disease Associations Subset of drug-function-pathway-disease network A. Predicted number of (A) KEGG pathways B. GO molecular functions affected by each drug MAPK network, many drugs target multiple disease classes, central pathway in pathogenesis of many diseases.

19 Experiment Validation I o Mebendazole binds with some kinases at affinity ranging from 5µM to 90nM o Mebendazole as a Therapeutic Option in Drug Resistant Melanoma in-vitro and in-vivo o Mebendazole program are planned for dose escalation clinical trial with new formulations 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article) 3. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

20 Experimental Validation I o As predicted by TMFS, Mebendazole inhibits several kinases at an affinity ranging from 15µM to 90nM VEGFR2 KIT PDGFRβ PDGFRα nm nm nm ABL1 DYRK1B JNK3 CDK7 P38α JNK2 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Naiem T, Oakland P, Byers S, Dakshanamurthy S. Comb. Chem. & High Thro. Scr (Most accessed article)

21 Experiment Validation II o Celecoxib (1.42 μm), and Dimethyl Celecoxib (1.56 μm) interacted with CDH11 o Celecoxib and DMC completed early stage preclinical studies in-vivo model 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

22 Celecoxib and DiMethyl Celecoxib binding to CDH11 (G) Gel Analysis: Celecoxib reduces the CDH11 dimer fraction (Native gel comparison of cadherin-11 in the absence (left) and presence (right) of celecoxib) (H) SPR assay: Celecoxib (1.42 μm), and Dimethyl Celecoxib (1.56 μm) interacted with CDH11 CDH11 CDH11 1. Sivanesan Dakshanamurthy et al. J. Med. Chem. 2012, 55, (triggered hundreds of press releases) 2. Assefnia et al. Cadherin-11 in poor prognosis malignancies and rheumatoid arthritis: common target, common therapies. Oncotarget Mar 30;5(6):

23 Summary TMFS method hit rate is greater than 91% accuracy for the majority of drugs. Anti-hookworm medication Mebendazole repurposed to specific kinases for possible drug resistant melanoma. Anti-inflammatory drug Celecoxib and DMC repurposed to CDH11, can be used for rheumatoid arthritis, IBD, and poor prognosis malignancies for which no targeted therapies exist. NextGenRePurposeVS provided new disease connections for many drugs and insight into their mechanism of action. Several drugs has been validated in-vitro and in-vivo, provided the validity of our screening tools; TMFS, RePurposeVS and NextGenRepurpose.