Nurix Therapeutics, Inc. · NASDAQ:NRIX
Nurix Therapeutics — 9th Annual TPD & Induced Proximity Summit (October 2026)
35 slides from Nurix Therapeutics's conference presentation, Oct 29, 2026
Slide 1: DEL AI: Proteome-Wide Binder Discovery with a DEL-Trained Foundation Model
DEL AI: Proteome-Wide Binder Discovery with a DEL- Trained Foundation Model Paul A Novick, Ph.D. 9th Annual TPD & Induced Proximity Summit October 29, 2026 Boston, MA
Slide 2: Industry Leading Discovery Engine for Induced Proximity Modalities
Industry Leading Discovery Engine for Induced Proximity Modalities Optimized DACs and Degraders Our in-house, integrated discovery engine leverages the combined power of our data-rich DEL capabilities, automated chemistry, and advanced machine learning to accelerate drug discovery 2 AI, artificial i
Slide 3: Induced Proximity Drug Discovery Requires Chemical Starting Points
Induced Proximity Drug Discovery Requires Chemical Starting Points Lead Identification is a central challenge in drug discovery 10,000× ~10B ~100M ~1M ~1K Fragment HTS Virtual screening Nurix DEL screening Library size by screening approach, compounds (log scale) 3 DEL, DNA-encoded library; HTS, hig
Slide 4: Scaffold-Based DELs Enable Hit Finding for Challenging Protein Surfaces
Scaffold-Based DELs Enable Hit Finding for Challenging Protein Surfaces >8B 52% Custom 3D scaffolds with high sp³ character allow better pocket fit unique compounds, built on complex, of ‘undruggable’ targets obtain hits custom 3D scaffolds solely from scaffold libraries DEL collection by library ty
Slide 5: DEL is a Platform That Delivers Chemical Assets for Traditionally Undruggable Targets
DEL is a Platform That Delivers Chemical Assets for Traditionally Undruggable Targets 2017 Today Targets with progressible* chemical matter Ligase Transcription Factor DUB Nuclear Receptor GPCR Target Class Kinase Epigenetic Enzyme Protease Fusion Approximate discovery timeline *All series validated
Slide 6: Nurix Generates Massive, High-Quality Datasets Ideal for Machine Learning
Nurix Generates Massive, High-Quality Datasets Ideal for Machine Learning High-Throughput Hit Selection & Target Protein DEL Screening Ligand Hit Confirmation Identification Production Acquisition DEL Screening Hit Selection Hit Confirmation 2,700+ 7T+ 900M+ 80+ DEL screens DEL datapoints Ligands ab
Slide 7: DEL Foundation Extends Discovery Beyond Physical Screening
DEL Foundation Extends Discovery Beyond Physical Screening and is the Key to Unlocking Induced Proximity Information-rich DEL screening datasets are ideally suited to advance machine learning 1,000× ~10T1 virtual ~10B+ ~100M ~1M ~1K Fragment HTS Virtual Nurix DEL DEL-AI screening screening Library s
Slide 8: DEL Foundation: A Generalized Model for Proteome-Wide Hit Identification
DEL Foundation: A Generalized Model for Proteome-Wide Hit Identification • Established DEL ML approaches train models on single target datasets • These models predict binding to a single protein and are inefficient to scale to ligand discovery across diverse targets • Rely on time and resource inten
Slide 9: DEL Foundation: A Generalized Model for Proteome-Wide Hit Identification
DEL Foundation: A Generalized Model for Proteome-Wide Hit Identification • Established DEL ML • DEL Foundation Model is approaches train models on trained on DEL datasets single target datasets from diverse protein targets • These models predict binding to a single protein • These models learn a and
Slide 10: DEL Foundation Model Architecture
DEL Foundation Model Architecture > 8×10⁹ Molecules Chemical CC(=O)O... FM CC(C)CC… CN1C=N… Protein-Ligand Neural Binding Network Prediction > 140 Targets Protein FM PLRSYC... GHDATN... MTKVQA... 10 FM, Foundation Model
Slide 11: DEL Foundation Dataset Construction and Definition of Terms
DEL Foundation Dataset Construction and Definition of Terms Experimental Actives Compounds which bound in the DEL screen 11
Slide 12: DEL Foundation Dataset Construction and Definition of Terms
DEL Foundation Dataset Construction and Definition of Terms Retrospective datasets constructed from experimental data Experimental Actives Experimental Inactives Compounds which bound in Compounds which did not bind the DEL screen in the DEL screen, sampled from full DEL library 12
Slide 13: DEL Foundation Dataset Construction and Definition of Terms
DEL Foundation Dataset Construction and Definition of Terms Retrospective datasets constructed Datasets not in the training dataset are evaluated from experimental data by DEL Foundation (DEL FM) Experimental Actives Experimental Inactives DEL FM Inactives DEL FM Actives Compounds which bound in Com
Slide 14: DEL FM Prospectively Predicts DEL Screening Results
DEL FM Prospectively Predicts DEL Screening Results Experimental Actives 14
Slide 15: DEL FM Prospectively Predicts DEL Screening Results
DEL FM Prospectively Predicts DEL Screening Results Experimental Actives DEL FM Scoring of Experimental Actives DEL FM Score -1 1 15
Slide 16: DEL FM Prospectively Predicts DEL Screening Results
DEL FM Prospectively Predicts DEL Screening Results Experimental Actives DEL FM Scoring of DEL FM Actives with Experimental Actives Experimental Annotation DEL FM Score Experimental Actives -1 1 Experimental Inactives 16
Slide 17: Rapid Improvement in Generalization to Novel Protein Space
Rapid Improvement in Generalization to Novel Protein Space First Generation DEL Foundation Model Highly Constrained by Sequence Similarity to Training Set 17 AUPRC, area under the precision-recall curve; DEL, DNA-encoded library; FM, foundation model
Slide 18: Rapid Improvement in Generalization to Novel Protein Space
Rapid Improvement in Generalization to Novel Protein Space First Generation DEL Foundation Model Highly Constrained by Sequence Similarity to Training Set 18 AUPRC Fold Above Random AUPRC, area under the precision-recall curve; DEL, DNA-encoded library; FM, foundation model
Slide 19: Rapid Improvement in Generalization to Novel Protein Space
Rapid Improvement in Generalization to Novel Protein Space Second Generation Model Outperforms First Generation Model On Low Identity Proteins 19 AUPRC Fold Above Random AUPRC, area under the precision-recall curve; DEL, DNA-encoded library; FM, foundation model
Slide 20: Broad Regions of the Proteome are Accessible with DEL Foundation Model
Broad Regions of the Proteome are Accessible with DEL Foundation Model By learning a general relationship between protein sequence and molecular structure, DEL FM transforms our existing wet-lab DEL data into a predictive model for hit and lead ID across a much larger portion of the proteome What we
Slide 21: Automated Prediction and Ligand Selection Pipeline Validated for Screening Unseen Targets
Automated Prediction and Ligand Selection Pipeline Validated for Screening Unseen Targets Protein Variant Target Protein Sequence Foundation Model Automated Ligand HTP On-Resin Prediction Selection Synthesis R O DEL FM N O Score Each Combination of Protein Sequence + N Enumerated Molecular H N Struc
Slide 22: Initial Prospective Test of First Generation Model: Kinase X Binders
Initial Prospective Test of First Generation Model: DEL Foundation Identified Diverse Set of High Affinity Binders for Kinase X Hit 1 20 0 Legend % Inhibition -20 Kinase Domain Di-synthon Followed Up -40 Sequence Validated hits -60 -80 -100 200µM IC50 1nM -120 DEL FM 10 -3 10 -2 10 -1 10 0 10 1 10 2
Slide 23: Second Test: Multiple Distinct Chemical Series Confirmed for TRIM15
Second Test: Multiple Distinct Chemical Series Confirmed From Initial Foundation Model Predictions for TRIM15 Hit 1 TRIM15 DEL-FM Predicted TRIM15 Hits 20 15 Response (RU) PRY/SPRY Domain BB2 Sequence 10 5 0 DEL FM 0 20 40 60 Prediction Time (s) Hit 2 TRIM15 20 Ligand 15 Response (RU) Selection 10 B
Slide 24: Third Test: Novel Series for An Additional Undrugged Ligase
Third Test: Novel Series for An Additional Undrugged Ligase Hit 1 20 WD40 Domain 15 Response (RU) Sequence 10 5 DEL FM 0 Prediction 0 20 40 60 Time (s) Hit 2 Ligand 20 Selection 15 Response (RU) 10 On-Resin 5 Resynthesis 0 0 20 40 60 Time (s) Hit 3 SPR 20 Validation 15 Response (RU) 10 Resample Lege
Slide 25: First Gen DEL Foundation Model Highly Productive in Prospective Screening
First Gen DEL Foundation Model Highly Productive in Prospective Screening First Generation DEL Foundation Model Has Yielded Binders for All Targets Evaluated to Date Kinase TRIM15 Ligase 1 Metric Value Metric Value Metric Value Compounds Tested 55 Compounds Tested 55 Compounds Tested 45 Confirmation
Slide 26: DEL FM Generalizes to New Chemical Space
DEL FM Generalizes to New Chemical Space UMAP of Nurix DELs in DEL FM Training Set UMAP_Y UMAP_X 26
Slide 27: DEL FM Generalizes to New Chemical Space
DEL FM Generalizes to New Chemical Space UMAP of Nurix DELs in DEL FM Training Set Schematic of Leave-One-Out Experiment 1 UMAP_Y 2 1 1 3 4 UMAP_X 27
Slide 28: DEL FM Generalizes to New Chemical Space
DEL FM Generalizes to New Chemical Space UMAP of Nurix DELs in DEL FM Training Set Schematic of Leave-One-Out Experiment 1 UMAP_Y 2 1 1 3 4 UMAP_X 28
Slide 29: DEL FM Generalizes to New Chemical Space
DEL FM Generalizes to New Chemical Space UMAP of Nurix DELs in DEL FM Training Set Schematic of Leave-One-Out Experiment 1 UMAP_Y 2 1 1 3 4 UMAP_X 29
Slide 30: DEL Foundation Model Can Also Be Employed for Hit Optimization
DEL Foundation Model Can Also Be Employed for Hit Optimization Protein Variant Target Protein Sequence Automated Ligand HTP On-Resin Foundation Model Selection Synthesis R O DEL FM N O Score Each Combination of Protein Sequence + N Enumerated Molecular H N Structure R every compound enumerated Confi
Slide 31: DEL FM Can Also Be Used to Optimize Model-Derived Hits
DEL FM Can Also Be Used to Optimize Model-Derived Hits: Suggests DEL Foundation Generalizes Beyond Training Data 10× affinity improvement 2× affinity improvement Selection Method Confirmation Rate DEL FM + Boltz 23% Diversity 10% 31 DEL, DNA-encoded library; FM, foundation model
Slide 32: Identification of Experimental Actives from DEL Adjacent Chemical Space
Identification of Experimental Actives from DEL Adjacent Chemical Space DEL FM predictions on our aggregated DEL Lead ID and Lead Optimization dataset cleanly separate confirmed binders from non-binding subsets Experimentally Tested DEL and DEL Adjacent Ligands Resynthesized ligands with no demonstr
Slide 33: Summary
Summary DEL FM is a powerful tool for drugging the proteome • Nurix’s DEL Foundation Model is trained on our aggregated DEL dataset covering diverse protein classes and proprietary library collection • DEL FM enables virtual DEL experiments which generalize to both new proteins and new chemical spac
Slide 34: Thank You to All Contributors Across Loka, Nurix, and AWS
Thank You to All Contributors Across Loka, Nurix, and AWS Machine Learning & Cristiana Carpinteiro Jennifer Gruefe Cheminformatics Sushmitha Regulapati Jim Davis Julian Fernandez Information Technology Teona Kostova Zafir Stojanovski Informatics Emily Kruger Maria Loureiro Protein Sciences Matheus D
Slide 35: NURIX
NURIX
Nurix Therapeutics — 9th Annual TPD & Induced Proximity Summit (October 2026) on 6ix