YOUR INNOVATIVE CHEMISTRY PARTNER IN DRUG DISCOVERY Design and Synthesis of the Comprehensive Fragment Library A 3D Enabled Library for Medicinal Chemistry Discovery Warren S Wade 1, Kuei-Lin Chang 1, Todd Meyer 1, Peter Pallai 1, Paolo Tosco 2, James Zapf 3, Laura Lingardo 3, Gordon Alton 3 1 BioBlocks, Inc., 2 Cresset, 3 Visionary Pharmaceuticals
Our Leap-to-Lead Platform Developed to Improve Lead Generation Leads for a current internal project Polarity Distribution Polarity Distribution Activity IC 50 Efficiency LE Activity Efficiency Size #HA Flexibility NRB Size Flexibility Polarity AlogP Polarity Literature Lead BioBlocks Early Lead Potent Not efficient Hydrophobic Not Permeable Failed in vivo Less Potent More efficient Balanced polarity Permeable ProperType plots illustrate Lead quality A good drug candidate hits the bullseye www.bioblocks.com 2
Enhancing Drug Discovery Leap-to-Lead Tools for property based optimization Comprehensive Fragment Library BioBlocks Proprietary Next Generation 3D Fragment Library Syntheverse Reaction-based access to currently synthesizable compounds www.bioblocks.com 3
Improved Fragment Screening Leap-to-Lead provides access to new chemical matter through its 3D enabled fragment screening set, the Comprehensive Fragment Library CFL designed for maximum diversity with medchem friendly properties Even commercial hits give access to novel, 3D structures Enables a head start on Hit to Lead and IP www.bioblocks.com 4
Building The Virtual Fragment Set An exhaustive set of ring structures containing 18 heavy atoms with 1 handle atom was built in BIOVIA Pipeline Pilot with several inclusion criteria: 1 ring, including bridged, spiro and fused ring connections 3 unique rings 2 rotatable bonds C,N,S,O atoms only 2 ring assemblies 1 atom has all rotatable bonds 1 S Rigid, partially aromatic structures are the highest value 3D enabling subset Core structures decorated with handles chosen for synthesis potential High Value Moderate Value Follow up only Not chosen www.bioblocks.com 5
Fragment Handle Criteria The raw set of >35 million structures with Me handles was filtered by medicinal chemistry criteria to produce a 7 million candidate fragment set Example CFL Fragments Excluded Reason Two handles* Unstable Isolated double bond* Too many rotatable bonds* *Potential follow up for a CFL fragment hit www.bioblocks.com 6
CFL Select set Practical computational limit is <1000k diastereomers Requires >600 CPU-years for full 3D similarity matrix A 580k subset was selected for clustering: Reduced mean of 16 heavy atoms to 14.6 Includes representatives of all 4500 ring types Includes 100% of <12 atom fragments to 1.5% of 18 atom fragments 2000000 1500000 Full set Selected 100% 80% 1000000 60% 40% 500000 20% 0 6 7 8 9 10 11 12 13 14 15 16 17 18 0% 6 8 10 12 14 16 18 www.bioblocks.com 7
Clustering Workflow Difficult examples Bridgehead Diastereomers Nonchiral atom Diastereomers Ring size limited Diastereomers Cis/trans ring Diastereomers Rule-based protonation Racemic diastereomer generation and 3D similarity based clustering Conformer and field point generation and Generation from 2D failed in difficult cases www.bioblocks.com 8
3D Clustering Alignment to methyl handle: Rotate structures around the handle Determine maximum similarity Similar compounds: Have common vectors Represent alternative sprouting choices Methyl handle Maximum similarity: 0.78 Comparison of two 6-member clusters Silhouette: 0.509 2D vs. 3D similarity 1 2D similarity 0.8 0.6 0.4 0.2 0 0 2D Cluster www.bioblocks.com 3D Cluster 9 0.5 1 3D similarity
1 2 3 4 6 10 16 25 40 63 100 158 251 398 631 1000 1584 3D Clustering Initial clusters from a full similarity matrix of 150k compounds Examined 10k, 25k and 50k clusters 14 CPU-years 10k gave a well-distributed 150k compound set; not scalable to the entire library 50k gave cluster sizes too small to be stable Remaining compounds placed in 25k clusters of most similar medoid PseudoSilhouette calculated because full matrix unavailable 20 CPU-years 1400 1200 1000 800 600 400 200 PseudoSilhouette*1000 Pop % Cumulative pop*10 7000 6000 5000 4000 3000 2000 1000 0-200 0 Mean intra-cluster similarity: 0.77 Binned Cluster Size, Mean 33 www.bioblocks.com 10
H Bonds are independent 2D clusters 2D similarity measure Same patterns are distributed over clusters ~50% of clusters have simple, valuable patterns Compounds prioritized by target interactions H Bonds important for: Fragment orientation Selectivity Fragments with 1 likely interaction motif preferred 51% 9% 40% DA-2 only DA-2 + others Not DA-2 www.bioblocks.com 11
HBond Patterns HBond patterns were generated for all compounds Pipeline Pilot Molecular Pharmacophore Fingerprints Component HBA, HBD, pairs when separated by 1-6 bonds 35 core patterns including Named Functional Groups Structure Pattern Name Pattern Frequencies DA-2 Amide Special case of DA-2 Urea DA-2,DA-2,DD-2 Pyridine Protonation state is handle dependent 180000 160000 140000 120000 100000 80000 60000 40000 20000 0 All 1 Feature 1 3 5 7 9 11131517192123252729313335 Pattern No. www.bioblocks.com 12
Picking Commercial Compounds Commercial Compounds found in 1395 clusters (5.6% of total) 905 clusters have only 1 commercial member, 81% are 3D enabled Others evenly distributed 2D and 3D by PMI* On rod/disk axis, 2 nd percentile of PMI area Examples 2D 3D 100 Members 90 Candidates 80 70 2D 60 50 40 30 20 10 0 * Metz, J.T. Principal Moments of Inertia Protocol, Pipeline Pilot Community. Available Cluster Distribution Total Candidates 2618 Picked 1085 Available 565 Clusters 310 Representative Clusters 11184 www.bioblocks.com 13
Building Plate 1 D A AA2 DA2 DD2 AA3 DA3 Vary cluster by well Vary Hydrogen bond patterns by row Vary core shape over column Vary Synthetic Accessibility by plate 29% Handle Type 3% 5% 2% 61% Me OH OMe X Other # Fragments MW AlogP HBD HBA Rot Bonds Unique Rings HBond Patterns % 3D % Commercial % Handles CFL Current Plates Average CFL Properties 96 1000 1700-30000 161 190 270 0.9 1.0 1.9 2.5 2-4 0.7 0.3 2-3 48 750 36 500 60 90 98 CFL Design Goals 0.8 1.1 64 30 98 Typical Commercial www.bioblocks.com 14
Library comparison CFL Current Plates Commercial Library 40% 30% 14% <25% Ar <75% Ar 33% 10% 25% <25% Ar <75% Ar 56% 60% <100% Ar % 3D % 2D 75% 57% <100% Ar % 3D % 2D www.bioblocks.com 15
Screening comparison Visionary Pharmaceuticals Same kinase target CFL Current Plate Commercial Library 56 4 14 9 13 IC50 Active Weakly Active Inactive Not Tested 127 416 68 38 IC50 Active Weakly Active Inactive Not Selective Compared with the previous library: Plate 1 has a lower hit rate, but more hits generate dose response curves More scaffold variation Allows follow-up on weak but interesting compounds Original Chemotypes still recognized www.bioblocks.com www.visionarypharmaceuticals.com 16
Connecting Fragment Hits to New IP Y/Z 1.0 Rods Principal Moments of Inertia Spheres 0.9 0.1 2 0.8 0.7 CFL000025 0.6 0.5 Discs 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 X/Z CFL000025 is a positive control for kinases 300 µm IC 50 Allows comparison of other hits to known hinge binding motif CFL provides 3D cluster of related compounds 47 member cluster Overlay of closest 20 shown Contains both known and novel alternatives Multiple 3D options www.bioblocks.com 17
Representative Hit Visionary Pharmaceuticals CFL clustering reveals many related compounds for rapid fragment SAR Independent Analogs CFL000025 14 Fragment Hits Handle Analogs 14 196 Handle Vectors 4 67 Same 2D Scaffold 37 2181 Same HBond Pattern 274 27370 Same 3D Cluster 47 1374 Total CFL Compounds 873 37984 % Unique to Compound 100% 80% CFL Analogs* >25,000 >450,000 *Estimated count of 3D clusters, all vectors www.bioblocks.com 18
Fragment Screening in Action Visionary Pharmaceuticals CFL provides 3D hits for a kinase target 25 20 15 10 5 0 7 8 9 10 11 12 13 14 15 16 17 18 Heavy Atoms Plate 1 Hits 60 50 40 30 20 10 0 Analogs Tested Analog Hits 7 8 9 10 11 12 13 14 15 16 17 18 Plate 1 Screened 14 hits, 36% 3D Mean IC 50 400 µm (50-2000) 3D Cluster Coverage 14 hit clusters, 83% 3D 2500 CFL structures Covers valuable 3D space CFL library follow up 35 hits Mean IC 50 220 µm (20 950) By minimizing overlap, a small library gave wide coverage of chemical space www.bioblocks.com www.visionarypharmaceuticals.com 19
Conclusion Available commercial compounds are disappointingly similar to each other. The CFL defines a valuable area of new chemistry space adjacent to known biologically active fragments. Our initial CFL screen generated high quality hits through its coverage of chemistry space. Enhanced 3D coverage revealed new, underexplored chemotypes useful for important drug targets. The CFL gives high quality entry points to FBLD such as our proprietary Leap-to-Lead discovery platform. www.bioblocks.com 20
Contact BioBlocks is developing Leap-to-Lead for use in Lead Discovery Collaborations. While the CFL is not available for independent purchase, we welcome new collaborators. For additional information or to discuss using the CFL and the Leapto-Lead platform in a drug discovery effort, please contact wwade@bioblocks.com or visit our website: www.bioblocks.com www.bioblocks.com 21