1. 1 HTS Follow-up using JChem Base: Virtual and Vendor Neighbors James Baxendale, Ajay, Keana Scott, Lalit Verma, Noel Southall, Trung Nguyen

2. 2 Problem – unintended consequences of HTS campaign success Chemists overrun with HTS hits to follow up – Workflow (GUI support) Visually identify SAR gaps Simplify access to gap filling sources (vendors, ideas, literature) Track progress in filling gaps Plot results to confirm / deny hypothesis – Technical (Under the covers with JChem Base) Capture and organize chemists follow up ideas Live test of virtual compounds against model Triage vendor compounds Pipeline to ease housekeeping, substructure search all vendors and manage vendor updates On-the-fly application of a large number of chemical and physical property filters Insert versus retrieval speed tradeoffs

3. 3 Workflow Solution – Visually Identify SAR Gaps Immediately see where activity (and other interesting properties) are conserved Zoom allows pattern detection for large results Visually construct a hypothesis then fill gaps to test

4. 4 Workflow Solution – Simplify access to gap filling sources SAR Matrix with Cells to fill highlighted and menu options shown Literature Compounds SDF VENDOR ASC results with exact Or near matches highlighted

5. 5 Workflow Solution – Track gap filling progress Share project level matrix overviews so everyone knows the status Request to fill gaps remembered and filled at later time automatically Overnight re-caching of results shows the latest results without the wait……… – Any new compounds, in-house synthesized, virtual or vendor? – Some that match existing gaps (or at least are close within some tolerance)? SEURAT Job Database Vendor Available Screening Compounds In House Synthesized Virtual Compounds Batch Exec of jobs

6. 6 Workflow Solution – Plot results to confirm / deny hypothesis Switch between scaffold specific SAR matrix and project level visualizations to determine whether further gap filling is warranted – Are potency and ADME / Tox getting better in recently synthesized compounds?

7. 7 Technical Solution - Capture and organize chemists follow up ideas Draw in from scratch, load an existing compound as a starting point or substructure search virtual compound space. Use JChem Base UpdateHandler to fill database fields for managing / tracking owner, date and status. Allow a view of synthesized and virtual side by side to aid idea generation

8. 8 Technical Solution - Live test of virtual compounds against model Hook into MarvinBeans mol property change event whenever molecule changes. Sum each non null contribution from a SMARTS match in the molecule, replace earlier matches with later ones Use JChem Base MolSearch.findAll to do SMARTS matching Use published journal articles to glean the appropriate SMARTS, their order and their contribution to each properties calculations. We are looking into addition of an logS computation among others

9. 9 Technical Solution - Triage Vendor Compounds 18 vendors were evaluated based upon library size, cost, availability, purity, analytical data and plating options. The collections from 8 vendors were further analyzed for duplicates: Test JChem CANSMI JChem CANSMI sufficient to use Drop vendors with lowest unique contribution

10. 10 Technical Solution – Pipeline to ease housekeeping, substructure search all vendors and manage vendor updates Pipeline to automatically process vendor updates upon SDF file arrival Keep track of duplicates as aliases but load only a single compound so we have options when ordering Normalize Names & Structures Calculate Physical Properties Calculate Chemical Properties and Encode Bitmask Property Rules (10) Chemistry Rules REOS (100) Multi-vendor DB Load JChem Base DB Using UpdateHandler Vendor SD Files Arrive AND Load support tables Using Hibernate Multi-vendor DB On average a file of 100,000 compounds takes ~3 hours to be processed

11. 11 Technical Solution – On-the-fly application of a large number of chemical and physical property filters 1.7 million compounds from 6 vendors (1.2 non-redundant) Main reasons for failing drug- like filters: – XLogP > 6.0 (213,308) – SSSR > 5 (45,363) – MW > 500 (126,346) – PSA > 160 (8217) – XLogP < -1.0 (4595) Passed Drug-like Filters 1,062,489 OR approx 75% Passed REOS Filters 814, 455 OR approx 57% Passed Drug-like AND REOS 629,611 OR approx 44% Allows searches on any combination of: – Sub-structure across any vendor combination – 10 drug like property ranges (e.g. MW, XLogP, PSA…) – 100 REOS flag maximums (e.g. Cyanidine Derivatives, alkene, amide, …) – Unique id or vendor name – Allowable elements soon

12. 12 Technical Solution - On-the-fly application of a large number of chemical and physical property filters Use positional bitmap for decent balance between insert and retrieval speeds otherwise an index would be needed on upward of 100 columns for decent performance (UTL_RAW for Oracle, PostgreSQL and others have equivalents) Bitwise OR can then check for a pass or fail of a concatenation of all flags at once 1 = 0001 2 = 0011 3 = 0111.. = …… 16 = …… Cyanidine Derivatives < 3 0111 Alkene < 1… 0001 … || = 01110001… PASS IF BIT_OR(concatFlagsThisQuery, precomputedInDB) = concatFlagsThisQuery AND

13. 13 Quick Demo & Contact Details Matrix SAR as the dashboard to manage gap filling Available Screening Compounds to show utility of on the fly queries even with a large number of chemical flags Q & A Visit out website for more detailed information about SEURAT – www.synapticscience.com – Pricing and sales information send a request to sales@synapticscience.com or call +1 301 915 0274 – SEURAT Components Available Separately SAR Tool Visualization plus generic spreadsheet Available Screening Compounds (ASC) Docking Viewer Kinase Viewer and Kinome Tree Viewer SMPC Tool (Hypothetical Compounds) – Other products MCS and detached MCS Patent space visualization SEURAT