Slide 1: Wrangling Diverse OmniAb Antibody Repertoires with OmniDeep
1 Wrangling Diverse OmniAb Antibody Repertoires with OmniDeep™ Bob Chen, PhD PEGS 2023
Slide 2: The OmniAb Technology Suite
2 The OmniAb Technology Suite The only platform leveraging four species Robust solutions for bispecific antibodies Human frameworks with ultralong CDR-H3s Most diverse host Antigen Design systems in the industry Proven success
Slide 3: Custom Antibody Repertoires for Every Target
3 Custom Antibody Repertoires for Every Target BIOLOGICAL INTELLIGENCE™: INTERPLAY BETWEEN RATIONAL GENETIC DESIGN AND POWERFUL IN VIVO PROCESSES Building Animal Systems Repertoire Shaping • Immunological robustness • Campaign strategy • V gene building blocks • Antigen design • Diversification arch
Slide 4: OmniAb Antibody Repertoires
4 OmniAb Antibody Repertoires BROAD PLATFORM AVAILABLE TO ADDRESS DIVERSE PARTNER OBJECTIVES Structural and Benefits for therapeutics Host V genes immunological features discovery and development • Full human V gene diversity • Diverse V gene usage and mixed • Widely accessible and flexible workflow
Slide 5: Common Light Chain Platforms
5 Common Light Chain Platforms STANDARD IGG FORMAT TO DE-RISK DOWNSTREAM DEVELOPMENT1 OF BISPECIFIC MABS Rearranged IgVK3-15/JK1 IgKC IgVK3-15/JK1 JK2-5 // Rearranged human VK3-15 light chain “Germlining” human VK3-15 light chain combined with diversifying heavy chain combined with diversifying heav
Slide 6: Discovery Platforms
6 Discovery Platforms Immunization Isolate B-cells Expression Validation Sequence Bioinformatics Hybridoma Workflow Fusion Screen B-cell enrichment Single-cell optional NGS Workflow Our powerful single B-cell screening technologies, xPloration® and GEM assay, bypass bottlenecks of hybridoma workflow
Slide 7: xPloration
7 xPloration® 1 | Loading 2 | Assay + Machine Vision 3 | Recovery & Single-Cell NGS Fluorescent Bright Field Antibodies 1.5 million, 40 µm 6 million, 20 µm Cell Surface Binding Label-free GFP Immune Reporter Activation cAMP Signaling 20 hr Primary Cells 24-72 hr Precise laser-based recovery Unique t
Slide 8: AI-Driven Multiplex Assays
8 AI-Driven Multiplex Assays • LandingLens AI platform enables selection of hits from xPloration based on multiple phenotypical assays Input Prediction Parameter 1 Parameter 2 Parameter 3 Parameter 1 Parameter 2 Parameter 3
Slide 9: Assay
9 Assay
Slide 10: Assay
10 Assay Target cell Antibody secreting cell
Slide 11: Assay
11 Assay
Slide 12: Laser Recovery
12 Laser Recovery
Slide 13: NKp46 Case Study
13 NKp46 Case Study: Discovering NK cell engager arm for bispecific antibody
Slide 14: Biological Background
14 Biological Background • Target ‒ NKp46 (NCR1, CD335) is a 46-kDa glycoprotein ‒ No statistically significant downregulation of NKp46 on both NK and T cells has been observed in many cancers • Goal: Discover anti-NKp46 antibodies from OmniClic and OmniFlic as one arm of a bispecific antibody (NKCE
Slide 15: Large-Scale Antibody Discovery
15 Large-Scale Antibody Discovery +
Slide 16: Multi-Parameter Screening: Multiplex Phenotypic Data
16 Multi-Parameter Screening: Multiplex Phenotypic Data Selective binding of target on beads BF Binding BF Binding Binding to NKp46 Binding to Tags Selective binding of target cell Negative Cell Binding Signal BF
Slide 17: Hit to Laser Recovery
17 Hit to Laser Recovery Target Negative Parental CHO K1 stained with CHO K1 NKp46 CellTracker Blue Red = Target 1x speed video Green = Negative
Slide 18: OmniFlic Screening Summary
18 OmniFlic Screening Summary Screen type # cells screened # hits Antigen on beads 5.0 M 1429 OABI01-1 Cells 7.7 M 345 Antigen on beads 3.7 M 751 OABI01-2 Cells 3.7 M 158 Antigen on beads 3.7 M 308 OABI01-3 Cells 3.7 M 33 • Recovered over 3000 positive events in 6 screens • Cells were extracted with
Slide 19: OmniFlic Repertoire Space
19 OmniFlic Repertoire Space Overlapping Overlapping # Unique Sequences # Lineages Lineages Unique Sequences OABI01-1 356 51 3 185 OABI01-2 621 42 4 445 OABI01-3 398 31 2 68 • 1375 Unique Clones in 124 lineages Protein + Protein Cell Cell binders binders • Focus on 9 overlapping lineages between cel
Slide 20: OmniFlic Repertoire Space
20 OmniFlic Repertoire Space Bead screen Cell screen
Slide 21: Bioinformatics-Aided Antibody Selection
21 Bioinformatics-Aided Antibody Selection • Clone selection considerations ‒ Sequence differences: maximize coverage of sequence space ‒ Sequence read counts: bias towards most represented via read count ‒ Bias towards or away lineage distribution All clones All clones Method A Method B Clone ID Cl
Slide 22: Structure-Based In Silico Developability Filter
22 Structure-Based In Silico Developability Filter VH & VL sequences Example 1: Fv_128 surface. Large lipophilic (yellow) surface CDR annotation & near the CDRs (Hydrophilic areas homology modeling shown in blue.) 3D homology models (Fvs) Static / dynamic (low mode MD) structure properties predictio
Slide 23: Most Selected Clones Pass In Silico Developability Filter
23 Most Selected Clones Pass In Silico Developability Filter Liability score (Goal = 0) • Total CDR Length • Area of Hydrophobicity near CDRs • VH-VL Charge balance Selected • Fv Charge Clinical Most selected OmniFlic antibodies pass • Filtered 2% of sequences CCG MOE implementation www.chemcomp.com
Slide 24: Discovery of High Affinity NKp46 Binders
24 Discovery of High Affinity NKp46 Binders # Selected Clones Expression (%) Binding (%) <10 nM (%) 301 95.7 84.7 20.1 Top 4 Lineages
Slide 25: OmniClic Screening Summary
25 OmniClic Screening Summary Screen Type # Cells Screened # Hits Antigen on beads 1.4 M 1200 69786 Cells 3.2 M 203 Antigen on beads 1.4 M 1199 69789 Cells 3.1 M 602 Antigen on beads 2.6 M 1326 69797 Cells 1.3 M 699 • Overall extraction count: 5229 • Overall unique sequences count: 2231
Slide 26: OmniClic Repertoire Space
26 OmniClic Repertoire Space Bead screen Cell screen
Slide 27: Parallel Processing of NGS and Cloning Workflows
27 Parallel Processing of NGS and Cloning Workflows In silico Expression Assay NGS Developability Validation # Speed xPloration Sanger Assay Cloning Expression Validation # Speed
Slide 28: Confirmed High Affinity Binders Overlay
Bead 28 Confirmed High Affinity Binders Overlay Cell Confirmed
Slide 29: OmniDeep: Deep Learning on Antibody Sequences
29 OmniDeep: Deep Learning on Antibody Sequences • Deep learning allows fuzziness and non-obviousness in antibody discovery and design ‒ Adaptable and scalable ‒ Help design and prioritization of novel antibodies with the desired therapeutic profile ‒ Data-driven: built on input data and/or associat
Slide 30: Encoding Antibody Repertoires into Latent Space
30 Encoding Antibody Repertoires into Latent Space The Decoder portion of the model also allows exploration of ^AMINOACIDSEQUENCE! sequences from the latent space Encoder (LSTM) Latent space trained with activity prediction is organized by activity Latent layer Dense layer Less active More active De
Slide 31: Scoring Model for Humanness
31 Scoring Model for Humanness • Humanness score – trained on human sequences from OAS database • Employs RoBERTa masked language Pseudo-LogLikelihoods (PLLs) from RoBERTa masked language model model ‒ robustly optimized bidirectional encoder Selected representations from transformers (BERT) Clinica
Slide 32: Validating VAE Latent Space
32 Validating VAE Latent Space Training Data Training Data + New Data (Masked) Training Data + New Data Principal Component 2 Principal Component 1 >1 μM 1 nM • Latent space is organized by affinity Affinity Unmeasured Data points from NGS • 2nd round of affinity data seems to follow latent space
Slide 33: Leveraging VAE Latent Space to Select Clones
33 Leveraging VAE Latent Space to Select Clones Use cases: 1. Guide to navigate antibody repertoire ‒ Focused on 5 high affinity clones ‒ Selected clones close in latent space but not in sequence • Non-obvious insights 2. Extend beyond observed sequences
Slide 34: VAE Highlight New Clusters to Prioritize
34 VAE Highlight New Clusters to Prioritize Bead Cell AI selected
Slide 35: Deep Repertoires Deep Sequencing Deep Screening Deep Learning
35 Deep Repertoires Deep Sequencing Deep Screening Deep Learning
Slide 36: OmniAb Technology Stack
OmniAb Technology Stack 36 Create Diverse Screen Antibody Identify the Antibody Repertoires Candidates Right Antibody Create Diverse Pools of High-Quality Screen Millions of Cells to Find Further Characterize, Select & Naturally Optimized Antibodies Potential Therapeutic Candidates Optimize the Righ
Slide 37: THANK YOU! Visit us at Booth #413
37 THANK YOU! Visit us at Booth #413 www.OmniAb.com